Donor Behaviour Analysis and Predictive Modelling for Blood Donation

Systems: A Data-Driven Approach in Nepal's Bagmati Region

Module Code & Module Title

Level 7 - FC7P01NI MSc Project

Assessment Weightage & Type

100% Individual Research

Year and Semester

2023-24 Spring-Autumn
Student Name: Sushil Dhital

London Met ID: 19031752

College ID: NP01MS7A220014

Assignment Due Date: 31st Jan 2025

Assignment Submission Date: 31st Jan 2025

Internal Supervisor Name: Mr. Aadesh Tandukar

External Supervisor Name: Dr. Sushil Shrestha

Word Count: 15184

I confirm that I understand my coursework needs to be submitted online via MySecondTeacher platform under the
relevant module page before the deadline in order for my assignment to be accepted and marked. I am fully aware
that late submissions will be treated as non-submission and a marks of zero will be awarded.
ORIGINALITY REPORT
ACKNOWLEDGEMENT

I extend my deepest gratitude to Dr. Sushil Shrestha, my external supervisor, for his invaluable
insights and expertise, which greatly enriched this research. I am equally thankful to my internal
supervisor, Mr. Adesh Tandukar, for his unwavering guidance and support throughout this study.
I am deeply grateful to Islington College for providing the academic platform and resources that
facilitated this research. Additionally, I wish to thank the Nepal Red Cross Society for providing
the dataset, which was instrumental in conducting this study. Their cooperation and support are
greatly appreciated.
ABSTRACT

Efficient blood donation management is vital for healthcare systems, particularly in resource-
constrained settings like Nepal. This study focuses on analyzing donor behavior and addressing
key challenges such as blood shortages and inefficient resource allocation through data-driven
strategies. Utilizing a dataset from the Nepal Red Cross Society, comprehensive data analysis was
conducted to identify demographic trends, geographic disparities, and donor behaviors. Predictive
modeling techniques, including XGBoost, were employed to forecast donation frequency and
donor engagement, while clustering analysis provided insights for targeted donor mobilization.

A prototype design for a mobile application was developed using iOS frameworks to demonstrate
potential features such as real-time notifications, donor matching, and geolocation services.
Although the backend implementation was not completed, the app design highlights a user-friendly
interface and features. A Streamlit dashboard was also created to visualize donor behavior insights
and predictive model outcomes effectively.

Furthermore, a survey conducted among licensed doctors validated the importance of data-driven
strategies and a dedicated mobile platform in addressing challenges such as donor availability,
emergency coordination, and resource allocation. The survey insights informed the design of the
app prototype and emphasized its relevance to healthcare providers.

This research contributes to understanding donor behavior and optimizing blood donation
management by leveraging advanced analytics, design prototyping, and stakeholder feedback. The
findings establish a foundation for future studies and practical applications aimed at improving
blood donation systems in Nepal.
TABLE OF CONTENTS

Chapter 1: Introduction ............................................................................................................... 1

1.1 Research Background ......................................................................................................................... 1

1.2 Research Gap ..................................................................................................................................... 1

1.3 Problem Statement ............................................................................................................................ 2

1.4 Research Questions............................................................................................................................ 3

1.5 Aims ................................................................................................................................................... 3

1.6 Objectives .......................................................................................................................................... 4

1.6.1 Data Analysis Objectives ............................................................................................................................ 4
1.6.2 Mobile Application Design Objectives........................................................................................................ 4
1.6.3 Alignment of Research Questions with Study Objectives .......................................................................... 5

1.7 Motivational Background ................................................................................................................... 6

1.7.1 Personal Motivation ................................................................................................................................... 6
1.7.2 Academic Relevance................................................................................................................................... 6

Chapter 2: Literature Review ...................................................................................................... 7

2.1 Overview of Blood Donation Management ....................................................................................... 7

2.2 Related Work and Existing Solutions ................................................................................................. 7

2.3 Identified Research Gap ..................................................................................................................... 9

2.4 How this thesis Addresses Research Gaps ....................................................................................... 10

Chapter 3: Theoretical Framework .......................................................................................... 11

3.1 Overview of Theoretical Framework................................................................................................ 11

3.2 Machine Learning Principles for Donor Analysis .............................................................................. 12

3.2.1 Supervised Learning ................................................................................................................................. 12
3.2.2 Clustering Models (Unsupervised Learning) ............................................................................................ 14

3.3 Statistical Foundations ..................................................................................................................... 15

3.3.1 Hypothesis Testing (T-Test, ANOVA) ........................................................................................................ 15
 3.3.2 Correlation Analysis and Insights ............................................................................................................. 17

3.4 Framework for Mobile Application Design ...................................................................................... 18

Chapter 4: Research Methodology ............................................................................................ 20

4.1 Structure of the Dataset................................................................................................................... 20

4.2 Data Analysis Process ....................................................................................................................... 21

4.2.1 Overview of Donor Dataset ...................................................................................................................... 22
4.2.2 Data Cleaning and Preprocessing ............................................................................................................. 27
4.2.3 Outlier Detection and Handling................................................................................................................ 31
4.2.4 Saving the Cleaned Dataset ...................................................................................................................... 33

4.3 Data Analysis Techniques ................................................................................................................. 36

4.3.1 Exploratory Data Analysis (EDA) ............................................................................................................... 36
[Link] Age, Gender, and Location Distributions .......................................................................................... 36
[Link] Blood Group Trends and Insights ..................................................................................................... 40
4.3.2 Statistical and Hypothesis Testing ............................................................................................................ 42
[Link] Correlation Matrix Analysis .............................................................................................................. 42
[Link] T-Test for Gender Based Differences................................................................................................ 44
[Link] ANOVA for Blood Group Comparisons ............................................................................................. 45
4.3.3 Predictive Modelling ................................................................................................................................ 47
[Link] Preparing the Dataset....................................................................................................................... 47
[Link] Linear Regression ............................................................................................................................. 48
[Link] Random Forest ................................................................................................................................. 50
[Link] Gradient Boosting ............................................................................................................................. 52
[Link] Support Vector Regression (SVR) ..................................................................................................... 54
[Link] XGBoost ............................................................................................................................................ 56
4.3.4 Clustering Analysis ................................................................................................................................... 58
[Link] Data Preparation for Clustering........................................................................................................ 58
[Link] Determining Optimal Number of Clusters ........................................................................................ 59
[Link] Applying K-Means Clustering ............................................................................................................ 60
[Link] Visualizing Clustering Results ........................................................................................................... 61

Chapter 5: System Prototype ..................................................................................................... 63

5.1 Survey Justifying the Need for the Application ................................................................................ 63

 5.2 Survey Results and Responses ......................................................................................................... 64

5.3 Overall Survey Results and Insights ................................................................................................. 70

5.3.1 Key Challenges Identified ......................................................................................................................... 70
5.3.2 Validation of Proposed Features .............................................................................................................. 70
5.3.3 Likelihood of Usage .................................................................................................................................. 70

5.4 Mobile Application (iOS) Name and Logo ........................................................................................ 71

5.5 Design and Features ......................................................................................................................... 72

5.5.1 Development Tools and Approach ........................................................................................................... 72
5.5.2 Core Features of the App (Design) ........................................................................................................... 72

5.6 Screenshots of User Interface .......................................................................................................... 74

5.7 Mobile App Limitations and Future Work ........................................................................................ 79

5.7.1 Current Limitation .................................................................................................................................... 79
5.7.2 Future Work ............................................................................................................................................. 79

Chapter 6: Results and Findings ............................................................................................... 80

6.1 Data Analysis Results ....................................................................................................................... 80

6.1.1 Exploratory Data Analysis (EDA) Results .................................................................................................. 80
6.1.2 Predictive Modelling Results .................................................................................................................... 80
6.1.3 Clustering Results ..................................................................................................................................... 81

6.2 Mobile Application Prototype Results ............................................................................................. 82

6.2.1 Survey Insights ......................................................................................................................................... 82
6.2.2 Prototype Evaluation................................................................................................................................ 82

6.3 Dashboard Implementation ............................................................................................................. 83

6.3.1 Exploratory Data Analysis Dashboard ...................................................................................................... 84
6.3.2 Predictive Modelling Dashboard .............................................................................................................. 85
6.3.3 Model Comparison Dashboard ................................................................................................................ 86
6.3.4 Clustering Dashboard ............................................................................................................................... 87

Chapter 7: Discussion ................................................................................................................. 88

7.1 Interpretation of Results .................................................................................................................. 88

7.2 Comparison with Related Works ..................................................................................................... 89

Chapter 8: Conclusion, Recommendations and Implication of Findings .............................. 90

Chapter 9: References ................................................................................................................ 91

Appendix ...................................................................................................................................... 95
TABLE OF FIGURES

Figure 1: Logo................................................................................................................................. 3
Figure 2: UIKit Framework (Kumar, 2024) ................................................................................. 18
Figure 3: Swift .............................................................................................................................. 19
Figure 4: XCode............................................................................................................................ 19
Figure 5: Signed Agreement for Data Collection ......................................................................... 20
Figure 6: Data Analysis Process (Dahal, 2021) ............................................................................ 21
Figure 7: Loading the Dataset ....................................................................................................... 22
Figure 8: Understanding the Dataset Structure ............................................................................. 23
Figure 9: Summary Statistics of the Dataset ................................................................................. 24
Figure 10: Missing Values in the Dataset ..................................................................................... 25
Figure 11: Data Types of the Dataset ........................................................................................... 26
Figure 12: Missing Values After Imputation ................................................................................ 27
Figure 13: Detecting and Handling Duplicates ............................................................................. 28
Figure 14: Validating Categorical Data ........................................................................................ 29
Figure 15: Validating Numerical Data .......................................................................................... 30
Figure 16: Boxplots for Age and Total Donations........................................................................ 31
Figure 17: Dataset Size Before and After Outlier Removal ......................................................... 32
Figure 18: Dataset Information After Cleaning ............................................................................ 33
Figure 19: Displaying First Five Rows of the Cleaned Dataset.................................................... 34
Figure 20: Saving the Cleaned Dataset ......................................................................................... 35
Figure 21: Screenshot showing saved file of clean dataset........................................................... 35
Figure 22: Age Distribution of Donors ......................................................................................... 37
Figure 23: Gender Distribution of Donors .................................................................................... 38
Figure 24: Location Distribution of Donors ................................................................................. 39
Figure 25: Bar Plot for Blood Group Distribution ........................................................................ 40
Figure 26: Percentage Distribution of Blood Groups ................................................................... 41
Figure 27: Correlation Matrix of Numerical Variables ................................................................ 42
Figure 28: Heatmap of Correlation Matrix ................................................................................... 43
Figure 29: Code to perform T-Test ............................................................................................... 44
Figure 30: T-Test Results for Gender-Based Differences in Donations ....................................... 45
Figure 31: Code to perform ANOVA Test ................................................................................... 46
Figure 32: ANOVA Test Results for Blood Group Comparisons ................................................ 46
Figure 33: Training and Testing Dataset Split Overview ............................................................. 47
Figure 34: Linear Regression Model Coefficients ........................................................................ 48
Figure 35: Evaluation Metrics for Linear Regression Model ....................................................... 49
Figure 36: Training of Random Forest Model .............................................................................. 50
Figure 37: Evaluation Metrics for Random Forest Model ............................................................ 51
Figure 38: Training of Gradient Boosting Model ......................................................................... 52
Figure 39: Evaluation Metrics for Gradient Boosting Model ....................................................... 53
Figure 40: Training of Support Vector Regression Model ........................................................... 54
Figure 41: Evaluation Metrics for Support Vector Regression Model ......................................... 55
Figure 42: Hyperparameter Tuning and Training of XGBoost Model ......................................... 56
Figure 43: Evaluation Metrics for XGBoost Model ..................................................................... 57
Figure 44: Data Preparation Summary for Clustering Analysis ................................................... 58
Figure 45: Code to Determine Optimal Number of Clusters ........................................................ 59
Figure 46: Elbow Curve for Determining Optimal Number of Clusters ...................................... 59
Figure 47: Clustering Results with Assigned Cluster Labels ....................................................... 60
Figure 48: Code to Visualize for Age vs Total Donations............................................................ 61
Figure 49: K-Means Clustering Visualization for Age vs Total Donations ................................. 62
Figure 50: Google Form Survey ................................................................................................... 63
Figure 51: Survey Question 1 ....................................................................................................... 64
Figure 52: Survey Question 2 ....................................................................................................... 65
Figure 53: Survey Question 3 ....................................................................................................... 65
Figure 54: Survey Question 4 ....................................................................................................... 66
Figure 55: Survey Question 5 ....................................................................................................... 66
Figure 56: Survey Question 6 ....................................................................................................... 67
Figure 57: Survey Question 7 ....................................................................................................... 67
Figure 58: Survey Question 8 ....................................................................................................... 68
Figure 59: Survey Question 9 ....................................................................................................... 68
Figure 60: Survey Question 10 ..................................................................................................... 69
Figure 61: Survey Question 11 ..................................................................................................... 69
Figure 62: iOS App Logo ............................................................................................................. 71
Figure 63: Login and Registration Pages ...................................................................................... 74
Figure 64: Dashboards of all three users....................................................................................... 75
Figure 65: Map Integration ........................................................................................................... 76
Figure 66: Alert Screens ............................................................................................................... 76
Figure 67: Profile Management Pages of all three users .............................................................. 77
Figure 68: Other Inner Screens ..................................................................................................... 78
Figure 69: Screenshot of Domain ................................................................................................. 82
Figure 70: Dashboard Implementation ......................................................................................... 83
Figure 71: Exploratory Data Analysis Dashboard ........................................................................ 84
Figure 72: Predictive Modelling Dashboard ................................................................................. 85
Figure 73: Model Comparison Dashboard.................................................................................... 86
Figure 74: Clustering Dashboard .................................................................................................. 87
TABLE OF TABLES

Table 1: Alignment of Research Questions with Study Objectives ................................................ 5

Table 2: A better understanding of the Dataset ............................................................................ 21
Table 3: Core Features of BloodConnect for Different Users ...................................................... 73
Table 4: Comparison of Predictive Models .................................................................................. 81
FC7P01NI MSc Project

Chapter 1: Introduction

1.1 Research Background

Blood donation plays a critical role in healthcare systems, providing life-saving support during
emergencies and routine medical procedures. However, in Nepal, the management of blood
donation faces numerous challenges, including frequent shortages that jeopardize patients' lives,
particularly during crises. These shortages are often exacerbated by inefficient communication
channels between blood donors, blood banks, and healthcare providers, leading to delays in
addressing urgent blood requirements.

Another significant issue is the lack of comprehensive data on donor demographics, such as age,
gender, and geographic distribution. This gap makes it difficult to develop targeted and effective
recruitment strategies. Despite global advancements in data-driven decision-making, Nepal’s
blood donation management systems have not yet fully embraced data analytics to optimize
operations and resource allocation.

This study aims to address these challenges by focusing on donor behaviour analysis and predictive
modelling to identify trends and optimize blood donation management. Additionally, a prototype
for a mobile application was designed to demonstrate features that could potentially enhance
communication, donor engagement, and resource coordination in the future. By leveraging data
analytics and machine learning techniques, this research provides insights into donor behaviour
and offers a foundation for developing more effective blood donation systems in Nepal.

1.2 Research Gap

Despite the vital role of blood donation in saving lives, Nepal's blood donation systems face
persistent challenges that hinder their efficiency. Current practices rely heavily on manual
processes and outdated methods, leading to delays in addressing urgent blood needs. The lack of
data-driven tools to analyse donation trends, predict shortages, and optimize resource allocation
further compounds these inefficiencies.

While global studies and initiatives have demonstrated the potential of data analytics and
predictive modelling in enhancing healthcare systems, there is limited research tailored to Nepal's

19031752 | Sushil Dhital | 1

FC7P01NI MSc Project

unique demographic and geographic context. Key challenges, such as the absence of a centralized
donor database and underutilization of advanced technologies, including real-time communication
tools and predictive models, remain unresolved. These gaps are particularly significant in the
Bagmati Region, which includes the capital Kathmandu and surrounding areas, where urbanization
and resource constraints create complex logistical challenges.

This study bridges these gaps by applying data analytics and predictive modelling to analyse donor
behaviour and identify actionable insights. It also incorporates the design of a prototype mobile
application to demonstrate how technology could enhance donor engagement and improve
operational efficiency. By focusing on a data-driven approach, this research contributes to the
growing need for innovative strategies to optimize blood donation management in Nepal.

1.3 Problem Statement

The management of blood donation systems in Nepal faces significant challenges that compromise
their efficiency and reliability. These challenges are driven by a lack of data-driven decision-
making, reliance on outdated communication methods, and the absence of technology integration.
Consequently, blood banks frequently struggle with shortages, particularly during emergencies,
leading to life-threatening delays in meeting critical blood demands.

Key Problems:

• Frequent Blood Shortages: Blood banks face critical shortages during emergencies due to
inadequate forecasting and the absence of effective donor mobilization strategies.
• Inefficient Communication Channels: Outdated communication methods between donors,
blood banks, and healthcare providers result in delays and hinder effective coordination.
• Absence of Centralized Donor Data: The lack of a unified database to track donor
demographics, donation history, and geographic patterns limits targeted recruitment and
engagement efforts.
• Limited Use of Predictive Tools: Current systems do not utilize data analytics to predict
shortages, analyse donation trends, or optimize resource allocation.
• Underutilization of Mobile Technology: Although mobile applications offer potential for
real-time notifications and donor engagement, their implementation remains limited in
Nepal.

19031752 | Sushil Dhital | 2

FC7P01NI MSc Project

1.4 Research Questions

This study aims to address the inefficiencies in Nepal's blood donation systems by leveraging data
analytics and predictive modelling to gain insights into donor behaviour and optimize resource
allocation. The following research questions guide the investigation and development of the
proposed solution:

• Research 1 (R1): What are the key demographic trends and geographic disparities in blood
donation behaviour within Nepal’s Bagmati region?
• Research 2 (R2): How can predictive and clustering strategies be used to analyse donor
behaviour, forecast donation frequency and optimize donor mobilization?
• Research 3 (R3): How can the integration of data-driven insights and technology
prototypes contribute to addressing key challenges in blood donation management?

1.5 Aims

The primary aim of this study is to address inefficiencies in Nepal's blood donation management
systems by leveraging data-driven approaches to analyse donor behaviour and demographic
patterns. This research seeks to optimize resource allocation, enhance donor mobilization
strategies, and provide actionable insights for improving blood donation systems in the Bagmati
Region. Additionally, a prototype for a mobile application has been designed to demonstrate
potential solutions for improving communication and donor engagement, setting the groundwork
for future technological advancements in this domain.

Figure 1: Logo

19031752 | Sushil Dhital | 3

FC7P01NI MSc Project

1.6 Objectives

The objectives of this study are divided into two main categories: data analysis and mobile
application development. These objectives collectively aim to address the identified challenges
and provide a comprehensive solution to enhance Nepal’s blood donation management and
emergency response systems.

1.6.1 Data Analysis Objectives

• To preprocess the donor dataset by addressing missing values, removing duplicates, and
handling outliers to ensure high-quality data for analysis. (O1)
• To conduct exploratory data analysis (EDA) for identifying key donor demographics,
geographic trends, and donation patterns. (O2)
• To apply predictive modelling techniques to analyse donation frequency and donor
behaviour trends. (O3)
• To segment donors into clusters based on age and donation frequency using clustering
algorithms for targeted donor mobilization strategies. (O4)
• To perform statistical analyses, including hypothesis testing and correlation analysis, to
understand relationships among variables and validate significant patterns. (O5)

1.6.2 Mobile Application Design Objectives

• To design a prototype for a mobile application that demonstrates potential features for
enhancing communication, donor engagement, and resource allocation. (O1)
• To illustrate features such as real-time notifications, donor matching, and geolocation
services based on insights from data analysis. (O2)
• To create an iOS-compatible user interface that showcases a user-friendly design,
emphasizing accessibility and usability for stakeholders. (O3)

19031752 | Sushil Dhital | 4

FC7P01NI MSc Project

1.6.3 Alignment of Research Questions with Study Objectives

Research Questions Objectives

RQ1: What are the key demographic • O1 (Data Analysis): Preprocess donor dataset for
trends and geographic disparities in demographic and geographic analysis.
blood donation behaviour within • O2 (Data Analysis): Conduct exploratory data analysis
Nepal’s Bagmati region? (EDA) for donor demographics and geographic trends.
• O4 (Data Analysis): Apply clustering algorithms to
identify regional disparities and donor behaviour patterns.
• O5 (Data Analysis): Perform statistical analyses to
validate significant trends and differences.
RQ2: How can predictive and • O3 (Data Analysis): Apply predictive models to analyse
clustering strategies be used to donation frequency and donor behaviour trends.
analyse donor behaviour, forecast • O4 (Data Analysis): Use clustering techniques to segment
donation frequency and optimize donors and develop targeted mobilization strategies.
donor mobilization?
RQ3: How can data-driven insights • O1 (Mobile Application Design): Design a prototype for
and a mobile application prototype real-time communication and donor engagement.
demonstrate solutions for improving • O2 (Mobile Application Design): Illustrate features such
donor engagement and resource as real-time notifications and geolocation-based donor
allocation in Nepal’s blood donation matching.
system? • O3 (Mobile Application Design): Ensure the prototype is
user-friendly and showcases potential functionalities
informed by data analysis.
Table 1: Alignment of Research Questions with Study Objectives

19031752 | Sushil Dhital | 5

FC7P01NI MSc Project

1.7 Motivational Background

The motivation for this research stems from both personal experiences and the academic drive to
address critical challenges in Nepal’s healthcare system. The frequent shortages and inefficiencies
in blood donation management systems, as highlighted in studies on public health in Nepal (NRCS,
2025) present an opportunity to make a meaningful impact by leveraging technology and data-
driven approaches.

1.7.1 Personal Motivation

My personal motivation for this research is rooted in witnessing the struggles faced by healthcare
providers in Nepal, particularly during emergencies, due to the lack of adequate blood resources.
Insights from my brother, who is a doctor, have further emphasized the difficulties in securing
timely blood donations for critically ill patients. These experiences have inspired me to contribute
to improving the healthcare system by applying data analytics and technological innovations to
address these pressing challenges. The aspiration to make a meaningful impact on public health
and address an issue that directly saves lives is a key driving force behind this research.

1.7.2 Academic Relevance

From an academic perspective, this study aligns with my specialization in data analytics and
project management, offering an opportunity to apply theoretical knowledge to a real-world
problem. Studies have shown that data-driven approaches, including machine learning and
exploratory data analysis, can transform healthcare systems in resource-constrained settings
(Epifani, et al., 2023); (Alkahtani & Jilani, 2019). This research not only integrates skills in
machine learning and mobile application design but also contributes to the discourse on leveraging
technology to optimize healthcare systems. By addressing a critical issue, this study not only adds
value to my academic journey but also aims to provide scalable, technology-driven solutions that
can serve as a model for similar challenges globally.

19031752 | Sushil Dhital | 6

FC7P01NI MSc Project

Chapter 2: Literature Review

2.1 Overview of Blood Donation Management

The procedure of collecting and supplying blood in Nepal is disorganised and poorly run. With
assistance from various organisations, blood banks that are permitted to run off-site voluntary
blood donation programs get blood from volunteer donors through donation drives. On-site
substitute donations are made by the patient's friends and relatives during blood shortages.
Following collection, the blood is analysed to ensure patient safety during transfusion. In
practically all hospitals, patients in need of blood and its products are in charge of making their
own arrangements for blood for medical procedures. This results in a blood search that is
disorganised, tense, and emotionally taxing, particularly in emergency situations. (Hamro
LifeBank, 2019)

2.2 Related Work and Existing Solutions

Blood donation management has been extensively studied, with a focus on predictive models,
donor profiling, and the use of machine learning techniques. A study by (Epifani, et al., 2023)
utilized Bayesian models to analyse patterns in blood donation frequency, successfully identifying
behavioural trends and improving donor arrival predictions. While this research provides valuable
insights into donor profiling, its application is limited to a single dataset, reducing geographic
diversity and scalability. Additionally, it lacks integration with real-time donor matching systems
or mobile platforms, leaving a gap in its practical implementation for emergency response.

(El-rashidy, et al., 2023) explored machine learning for classifying donors based on their
likelihood to donate. This study effectively identified predictive features and achieved accurate
donor classification using techniques like decision trees and random forests. However, it was
limited to classification tasks, without expanding into clustering or regression models. The dataset
used lacked diversity, focusing on a single region, and the research did not address real-time
usability or emergency-specific applications, which are critical for practical blood donation
systems.

Another study by (Selvaraj, et al., 2022) applied decision trees and support vector machines
(SVMs) to predict donor return behaviour. While the research provided a solid foundation for

19031752 | Sushil Dhital | 7

FC7P01NI MSc Project

predictive analytics, it failed to incorporate features for dynamic donor engagement or integration
with emergency response systems. The dataset, sourced from a Saudi blood bank, was region-
specific, restricting the generalizability and scalability of its findings. Moreover, the study did not
address the potential use of geolocation-based donor tracking or real-time notifications, limiting
its real-world applicability.

(Alkahtani & Jilani, 2019) focused on time series analysis to predict the likelihood of donor
returns over time. Their research successfully identified trends in donor behaviour and developed
a predictive model for return rates. However, the study lacked scalability and practical integration
into real-time systems or mobile platforms. The dataset was again region-specific, and the research
did not explore how these predictions could inform emergency response strategies or donor
mobilization efforts.

These studies collectively highlight the progress made in predictive analytics and donor profiling
for blood donation management but also underline significant gaps. Key areas for improvement
include the integration of real-time systems, the use of diverse and scalable datasets, and the
application of predictive insights to emergency response and donor engagement. This study aims
to address these gaps by leveraging data analytics, predictive modelling, and clustering techniques,
while proposing a prototype mobile application as a foundation for future technological
advancements in Nepal's blood donation systems.

19031752 | Sushil Dhital | 8

FC7P01NI MSc Project

2.3 Identified Research Gap

The analysis of the reviewed research papers reveals several critical gaps in the current body of
work on blood donation management and its integration with emergency response systems. A
recurring limitation is the geographic and dataset constraints, as many studies focus on specific
regions, such as AVIS Milan (Epifani, et al., 2023) or Saudi blood banks (Selvaraj, et al., 2022),
(Alkahtani & Jilani, 2019), with limited demographic diversity and scope. This restricts the
generalizability and scalability of their findings. Additionally, while predictive modelling and
classification techniques are commonly employed, these approaches often lack real-time
applications. For instance, donor profiling (Epifani, et al., 2023) and time series analyses
(Alkahtani & Jilani, 2019) are not connected to dynamic donor-recipient matching or emergency
mobilization scenarios. Furthermore, the narrow focus on specific machine learning approaches,
such as classification or decision trees (El-rashidy, et al., 2023), (Selvaraj, et al., 2022) neglects
the potential of clustering, regression, or ensemble methods to enhance predictive accuracy and
insights.

Another significant gap lies in the limited integration of mobile technology to operationalize these
analytical findings. While donor profiling and predictive modelling offer valuable insights, they
are not linked with geolocation-based systems or mobile applications to facilitate donor
engagement and resource coordination (Selvaraj, et al., 2022), (Alkahtani & Jilani, 2019). Lastly,
most studies fail to validate their models through practical implementation, and scalability to
diverse geographic or socio-economic contexts remains unaddressed. These gaps highlight the
need for a data-driven approach that integrates predictive modelling, clustering, and geolocation-
based technologies while also addressing the scalability and real-world applicability of blood
donation systems. This thesis aims to bridge these gaps by leveraging diverse datasets and
proposing solutions tailored to the specific needs of Nepal's Bagmati Region.

19031752 | Sushil Dhital | 9

FC7P01NI MSc Project

2.4 How this thesis Addresses Research Gaps

This thesis addresses the identified research gaps by leveraging data-driven approaches to improve
blood donation management systems in Nepal's Bagmati Region. Through comprehensive data
analysis, this study examines donor demographics and geographic disparities to provide actionable
insights for targeted donor engagement and mobilization. Predictive modeling techniques, such as
regression and clustering, are utilized to analyze donation frequency and segment donors, enabling
more efficient strategies for resource allocation. These approaches fill gaps in the literature where
prior studies focused narrowly on classification tasks or lacked integration with diverse datasets
and broader methodologies.

Although the mobile application remains a prototype, it plays a key role in demonstrating the
potential for real-time donor engagement and geolocation-based donor matching. The design
integrates insights from data analysis to illustrate features like dynamic notifications and
emergency alerts, addressing gaps in real-time applications highlighted in the literature.
Furthermore, this research emphasizes scalability and practical implementation by analyzing a
dataset that incorporates diverse demographic and geographic contexts, ensuring broader
applicability beyond a single region. By combining donor profiling, predictive modeling, and the
conceptualization of real-time mobilization tools, this thesis provides a foundation for innovative
and scalable solutions to the challenges in blood donation management. Ultimately, this research
contributes to optimizing donor engagement and resource allocation while laying the groundwork
for future advancements in data-driven healthcare systems.

19031752 | Sushil Dhital | 10

FC7P01NI MSc Project

Chapter 3: Theoretical Framework

The theoretical framework serves as the foundation that supports the theory behind a research
study. It provides a structured approach, guiding the organization of ideas, the development of
research questions, and the flow of discussions, conclusions, and knowledge transfer. Essentially,
it shapes the overall structure of the research, ensuring coherence and alignment with the study's
objectives. A strong theoretical framework incorporates key elements such as methodology (the
approach to knowing), ontology (the understanding of existence), and epistemology (the study of
knowledge). Without these components, a theory cannot progress or be effectively applied in
research. (Brodowicz, 2024)

3.1 Overview of Theoretical Framework

The foundation of this research is built on the integration of data analytics, machine learning, and
mobile application development to tackle the pressing issues in Nepal’s blood donation
management and emergency response systems. These frameworks shape the methodologies used
in the study, ensuring that they align with the objectives and research questions. Machine learning
techniques are applied to create predictive models that analyse donor behaviour and anticipate
blood shortages, helping to allocate resources more effectively. In addition, statistical tools like
exploratory data analysis (EDA) are used to uncover patterns in donor demographics, geographic
disparities, and donation trends, providing critical insights for targeted donor engagement
strategies.

The use of mobile application frameworks complements the data analysis by proposing a platform
that facilitates real-time communication and better donor involvement. This conceptual app design
aims to bridge the existing gaps in coordination between donors, blood banks, and healthcare
providers, addressing the inefficiencies of current systems. Together, these frameworks create a
solid foundation for the research, leveraging innovative, data-driven approaches to improve the
efficiency and communication of Nepal’s blood donation management processes. This holistic
approach not only addresses existing challenges but also highlights the potential of combining
advanced analytics and technology to drive meaningful improvements in healthcare systems.

19031752 | Sushil Dhital | 11

FC7P01NI MSc Project

3.2 Machine Learning Principles for Donor Analysis

Building models that can learn from data and make predictions or judgements without explicit
programming is at the heart of machine learning (ML) ideas. These guidelines guarantee that
machine learning (ML) systems are efficient, dependable, and able to generalise to fresh data.
(vaibhav_tyagi, 2024)

Machine learning plays a vital role in this research, offering powerful tools to analyse donor data
and uncover meaningful patterns. By leveraging supervised learning techniques, this study
develops predictive models that provide insights into donor behaviour and trends. These models
are designed to address inefficiencies in Nepal's blood donation management system, helping to
optimize resource allocation and improve donor mobilization strategies. Through this approach,
machine learning not only supports the study’s objectives but also demonstrates its potential to
bring impactful, data-driven solutions to real-world challenges.

3.2.1 Supervised Learning

Supervised learning is a kind of machine learning that trains computers to identify patterns and
predict outcomes using labelled information. (Google Cloud, 2024) In this study, supervised
learning was applied to predict Total Donations using features such as Age, Gender, Location, and
Blood Group. Key algorithms used include:

• Linear Regression:

In this study, Linear Regression is utilized as a foundational model to predict total donations
by exploring the linear relationships between the variables and the target outcome. Its
simplicity and ease of interpretation make it an ideal starting point for understanding how
different factors influence donor behavior. However, since Linear Regression is limited to
capturing linear patterns, it may not fully account for the complexities and nuances inherent in
donor dynamics.

Here, Linear Regression serves as a benchmark against which more advanced models are
compared. By offering an initial understanding of the data and revealing basic patterns, it
provides valuable insights that pave the way for developing more sophisticated predictive

19031752 | Sushil Dhital | 12

FC7P01NI MSc Project

methods. Its straightforward approach establishes a baseline for performance evaluation,

setting the stage for refining predictions using more advanced machine learning techniques.

• Random Forest Regressor:

The Random Forest Regressor is used in this study to manage non-linear relationships and
interactions between variables, making it a reliable choice for modeling complex patterns in
the data. Its strength lies in combining multiple decision trees, which enhances predictive
accuracy and ensures stability. Another advantage of Random Forest is its ability to highlight
the importance of different features, offering insights into which factors most significantly
influence total donations. Its flexibility in handling mixed types of data adds to its suitability
for diverse datasets.

In this research, Random Forest is applied to overcome the limitations of simpler models like
Linear Regression, capturing more intricate relationships and interactions within the dataset.
By identifying the key factors driving donation behavior, it not only boosts prediction accuracy
but also provides meaningful insights to design targeted donor engagement strategies. This
approach aligns closely with the study's objective of improving resource allocation and
strengthening donor mobilization through data-driven methodologies.

• Gradient Boosting Models (GBM):

Gradient Boosting is applied in this study as a powerful method to enhance prediction accuracy
by iteratively combining multiple weak models and correcting their errors step by step. This
sequential approach allows it to effectively handle complex data and capture intricate patterns
that simpler methods might miss. In many cases, Gradient Boosting has the potential to
outperform models like Random Forest by leveraging its error-focused refinement process to
deliver more accurate predictions.

In this research, Gradient Boosting plays a key role in improving the precision of Total
Donation predictions. By systematically addressing the limitations of earlier models, it
incrementally enhances performance, resulting in more reliable and detailed insights. This
aligns with the study's goal of employing advanced machine learning techniques to better
allocate resources and engage donors. Its adaptability and ability to learn from previous

19031752 | Sushil Dhital | 13

FC7P01NI MSc Project

mistakes make Gradient Boosting an essential component of the research, especially for
navigating the complexities of blood donation data.

• XGBoost Regressor:

The XGBoost Regressor is utilized in this study to optimize prediction performance through a
scalable and regularized boosting algorithm. Known for its efficiency and adaptability,
XGBoost excels in handling structured data and reducing overfitting, making it a powerful tool
for complex datasets. Additionally, it provides feature importance metrics, offering valuable
insights into the factors driving model decisions and predictions.

In this thesis, XGBoost is chosen for its speed and superior performance, particularly in
scenarios with limited resources. Its ability to fine-tune predictions and effectively manage
complex datasets aligns with the study's objectives of accurately analyzing donor behavior and
optimizing total donation predictions. By leveraging its strengths, XGBoost contributes
significantly to achieving precise and actionable insights in the context of blood donation
management.

3.2.2 Clustering Models (Unsupervised Learning)

Clustering is an unsupervised learning technique that groups data points based on similarities. In
this study, K-Means clustering was used to segment donors into distinct groups, enabling targeted
donor engagement strategies.

• K-Means Clustering:

K-Means clustering was used in this study to segment donors into distinct groups, providing
valuable insights for designing targeted donor engagement strategies. By identifying common
characteristics within each cluster, this approach offers a deeper understanding of donor
behavior and helps in crafting more effective outreach efforts.

The goal of applying K-Means clustering was to group donors based on their age and total
donation frequency, uncovering meaningful patterns in their behavior. Using the Elbow
Method, the optimal number of clusters was determined to be four. These clusters included
young, low-frequency donors (Cluster 0), moderate-age donors with moderate donation
frequencies (Cluster 1), diverse-aged, high-frequency donors (Cluster 2), and older, consistent

19031752 | Sushil Dhital | 14

FC7P01NI MSc Project

donors (Cluster 3). K-Means was selected for its straightforward yet effective approach to
segmenting the dataset, allowing for actionable insights that can enhance donor mobilization
and optimize resource allocation.

3.3 Statistical Foundations

Statistical methods are an essential part of this research, offering powerful tools to validate patterns
in donor data and explore meaningful relationships between variables. These techniques help
provide a clearer understanding of trends and behaviours within the dataset, enabling precise
hypothesis testing and drawing reliable conclusions. Methods like T-tests, ANOVA, and
correlation analysis were employed to ensure that the study’s findings are backed by thorough and
accurate analysis.

By incorporating these statistical approaches, the research not only strengthens the credibility of
its results but also translates them into actionable insights. This supports informed decision-making
in areas such as donor mobilization and resource allocation. These methods form a robust
foundation for addressing key challenges in blood donation management, paving the way for a
more efficient, data-driven approach to improving healthcare systems.

3.3.1 Hypothesis Testing (T-Test, ANOVA)

Hypothesis testing evaluates whether observed differences in data are statistically significant or
occur by chance.

• T-Test:

The T-Test is a statistical method used to compare the means of two groups and determine
whether the difference between them is statistically significant. This test is particularly useful
for binary comparisons, such as evaluating differences in donation behaviours between male
and female donors. By focusing on group-level differences, the T-Test provides a
straightforward yet powerful way to identify patterns within the dataset, offering insights into
how specific variables influence donor behaviour.

In this thesis, the T-Test is employed to uncover demographic trends and guide targeted
interventions for donor engagement. By identifying statistically significant differences

19031752 | Sushil Dhital | 15

FC7P01NI MSc Project

between groups, such as gender-based variations in donation behaviour, this analysis helps
inform strategies to tailor communication and outreach efforts. The use of the T-Test ensures
that the study’s conclusions are grounded in rigorous statistical analysis, supporting its
objective to optimize donor mobilization and improve resource allocation.

• ANOVA (Analysis of Variance):

ANOVA, or Analysis of Variance, is a statistical technique used to assess whether the means
of three or more groups differ significantly from one another. This method is especially useful
for identifying variations across categories, such as differences in donation frequencies among
various blood groups. By examining these group-level differences, ANOVA helps uncover
meaningful patterns and disparities within the data, making it a powerful tool for exploring and
understanding complex relationships.

In this thesis, ANOVA is applied to uncover meaningful variations in donation behaviour

across multiple groups, such as blood types, to identify actionable trends. This analysis informs
policy and strategy by highlighting areas where targeted interventions may be needed to
address disparities or optimize donor engagement. By incorporating ANOVA into the research
framework, the study ensures a data-driven approach to understanding group-level variations,
supporting its objective to enhance resource allocation and donor mobilization efforts
effectively.

19031752 | Sushil Dhital | 16

FC7P01NI MSc Project

3.3.2 Correlation Analysis and Insights

Correlation analysis measures the strength and direction of relationships between variables. In this
study:

• Age and Total Donations:

Correlation analysis is used to examine the relationship between age and total donations,
measuring how donor age influences their donation behaviour. This analysis helps uncover the
strength and direction of the relationship, providing valuable insights into whether certain age
groups are more likely to donate frequently. By focusing on age as a demographic variable,
this method allows for a deeper understanding of how life stages might affect donation patterns.

In this thesis, the relationship between age and total donations is analysed to determine if age
is a significant predictor of donation frequency. These insights are essential for tailoring donor
mobilization strategies to different age groups. For instance, if younger donors are less frequent
contributors, targeted outreach campaigns could be developed to increase their participation.
This approach ensures that donor engagement strategies are informed by data, enhancing their
effectiveness.

• Blood Pressure and Total Donations:

Correlation analysis is also applied to evaluate the relationship between physical health
indicators, such as blood pressure, and donation behaviour. This analysis investigates whether
factors like blood pressure have a measurable impact on the frequency of blood donations,
providing a holistic view of how physical health might influence donor behaviour. Even weak
or non-existent relationships are valuable in refining the scope of analysis and understanding
the dataset comprehensively.

In this thesis, analysing blood pressure and donation frequency ensures that all potential factors
influencing donation behaviour are considered. While the relationship may not always be
strong, including such variables demonstrates a commitment to comprehensive analysis and
highlights any overlooked trends. These findings contribute to building a complete profile of
donor behaviour, aligning with the study’s objective of leveraging data-driven insights to
optimize blood donation management.

19031752 | Sushil Dhital | 17

FC7P01NI MSc Project

3.4 Framework for Mobile Application Design

The development of the mobile application was structured to follow a systematic framework,
ensuring clarity, efficiency, and alignment with the project’s objectives. The framework was
designed to incorporate an incremental approach, focusing on iterative progress and continuous
feedback.

Figure 2: UIKit Framework (Kumar, 2024)

The incremental model is a software development approach that builds systems step by step by
adding new features or "increments" based on an analysis of the source data. Rather than starting
from scratch, this model enhances an existing system by dividing software requirements into
smaller, manageable modules, each representing a stage in the software development life cycle
(SDLC). These stages include testing, design, implementation, analysis, and maintenance, where
each completed module lays the foundation for the next. This iterative process ensures that the
functionality improves incrementally until the entire system is complete. (Plutora, 2023)

In this study, although only the app design phase was completed, the incremental methodology
provided a structured approach to prioritizing and developing features. The design focused on
creating intuitive user interfaces and conceptualizing database structures to manage donor
information and blood inventory. Feedback from stakeholders was integral to refining the
prototype, ensuring that the design aligned with user needs and the research objectives. This
phased approach highlighted the app’s potential to enhance donor engagement and
communication, offering a solid starting point for further development and eventual
implementation. (Plutora, 2023)

Although only the app design was completed, the incremental methodology ensured a structured
approach to feature prioritization and development. The design process involved creating intuitive

19031752 | Sushil Dhital | 18

FC7P01NI MSc Project

user interfaces and conceptualizing database structures to store donor information and blood
inventory details. Stakeholder feedback played a crucial role in refining the prototype, ensuring it
met user needs and aligned with the study's objectives. This phased approach demonstrated the
potential of the mobile application to enhance donor engagement and communication, setting a
foundation for future development and implementation. (Plutora, 2023)

The mobile application was developed specifically for iOS devices, leveraging technologies and
tools optimized for the platform. Key considerations for iOS development included the use of
Swift as the programming language and Xcode as the integrated development environment (IDE).

• Swift Programming Language

Swift is a powerful and intuitive programming language designed to work

seamlessly with all Apple products. Its modern features and concise yet
expressive syntax make it approachable, even for those new to programming.
Swift prioritizes both performance and safety, enabling developers to create
software that runs incredibly fast while minimizing common errors. This
Figure 3: Swift
combination of speed, reliability, and ease of use has made Swift a preferred
choice for developing robust applications in the Apple ecosystem. (Apple, 2025)

• Xcode IDE

Xcode is a comprehensive development environment that allows

developers to create, test, and distribute apps across all Apple
platforms. With features like interactive previews, live animations,
and predictive code completion, Xcode streamlines the app
development process, making it faster and more efficient.
Developers can manage their code seamlessly using Git staging,
enabling them to prepare commits without disrupting their
Figure 4: XCode
workflow. Enhanced test reports, complete with video recordings,
provide valuable insights for analysing and improving test results. Additionally, Xcode Cloud
simplifies deployment, allowing apps to be effortlessly distributed to TestFlight and the App
Store. These features make Xcode an indispensable tool for building exceptional apps within
the Apple ecosystem. (Apple, 2025)

19031752 | Sushil Dhital | 19

FC7P01NI MSc Project

Chapter 4: Research Methodology

This chapter outlines the methods, tools, and processes used to conduct data analysis and achieve
the research objectives. It includes detailed steps, from data collection and preprocessing to
applying analytical techniques and project management practices. The focus is on using Python
for data-driven insights and predictive modelling, ensuring reproducibility and transparency
throughout the process.

4.1 Structure of the Dataset

The dataset used in this study was collected from the Nepal Red
Cross Society (NRCS), which is the primary organization
managing blood donation activities in Nepal. This dataset
provides critical insights into donor demographics, donation
trends, and regional disparities, serving as a foundation for
data-driven analysis and model development. The dataset has
been carefully reviewed and prepared to ensure its suitability
for addressing the research objectives. Ethical considerations
were prioritized, with informed consent ensured for all
participants. Data privacy was safeguarded under NRCS
guidelines, and institutional ethical approval was obtained
before analysis. Figure 5: Signed Agreement for Data Collection

The dataset contains 5,123 observations and 6 variables, each contributing vital information for
analysing blood donation patterns and donor characteristics. The featured variables and their
descriptions are as follows:

Data Frame Name Variable Type Variable Description

Location Categorical Shows the geographical location of the donor.

Age Numerical Shows the age of the donor.

Gender Categorical Specifies the gender of the donor.

19031752 | Sushil Dhital | 20

FC7P01NI MSc Project

Blood Group Categorical Indicates the blood group of the donor.

Blood Pressure Numerical Shows the blood pressure level of the donor.

Total Donations Discrete Indicates the total number of donations made

by the donor.
Table 2: A better understanding of the Dataset

4.2 Data Analysis Process

Data analysis is a systematic process that involves multiple steps to prepare, examine, and interpret
datasets. The goal is to ensure the data is clean, accurate, and ready for statistical or predictive
analysis. This process includes examining, cleaning, transforming, and modelling data to uncover
meaningful insights, draw conclusions, and support decision-making. It often requires working
with data from diverse sources and formats, both structured and unstructured, using various
techniques and strategies to make sense of complex information. (Crabtree & Nehme, 2024)

• Data Overview and Loading

• Data Cleaning and Preprocessing
• Outlier Detection and Handling
• Data Summary

Figure 6: Data Analysis Process (Dahal, 2021)

19031752 | Sushil Dhital | 21

FC7P01NI MSc Project

4.2.1 Overview of Donor Dataset

This section involves loading the dataset and inspecting its structure to gain an initial
understanding of the data. The following steps will be conducted:

Objective: Load the dataset into Python and verify that it is correctly imported for further analysis.

Step 1: Loading the Dataset

The first step involves importing the dataset into Python to ensure it is ready for analysis. This step
focuses on verifying that the data is correctly loaded.

Figure 7: Loading the Dataset

Observations:

The dataset contains records about blood donors, including:

- Columns: locations, age, blood groups, genders, blood pressure, and the total number of
donations.
- Initial Verification: Data successfully loaded, displaying the first few rows to confirm the
structure.

19031752 | Sushil Dhital | 22

FC7P01NI MSc Project

Step 2: Understanding the Dataset Structure

To ensure the dataset was properly loaded, its structure was inspected using the info() method in
Python. This provided a comprehensive summary of the dataset, including the column names, data
types, non-null counts, and memory usage.

Figure 8: Understanding the Dataset Structure

Observations:

• The dataset contains a total of 5,123 entries and 6 columns.

• All columns, except Age, have complete data (no missing values), while Age has 3 missing
entries.
• The dataset appears to have a balanced structure, providing essential information for
analysis.

19031752 | Sushil Dhital | 23

FC7P01NI MSc Project

Step 3: Displaying Summary Statistics

To generate and inspect summary statistics of the numerical columns in the dataset to understand
their range, central tendencies, and variability.

Figure 9: Summary Statistics of the Dataset

Observations:

• The Age column has:

- A mean age of approximately 36.54 years.
- A minimum age of 18 years and a maximum age of 69 years.
- The standard deviation is 12.03, indicating some variability in donor ages.
• The Total Donations column shows:
- A mean of 7.27 donations, with a range from 1 to 104 donations.
- The standard deviation is 8.74, suggesting high variability in donation frequencies.
- 75% of donors have made up to 10 donations, while a few outliers have exceptionally
high donation counts.
• The dataset exhibits consistent numerical ranges suitable for analysis.

19031752 | Sushil Dhital | 24

FC7P01NI MSc Project

Step 4: Check for missing values

To identify and quantify missing values in each column of the dataset, ensuring data completeness
before proceeding to cleaning and preprocessing.

Figure 10: Missing Values in the Dataset

Observations:

• The dataset has 3 missing values in the Age column.

• All other columns, including Gender, Location, Blood Group, Blood Pressure, and Total
Donations, have no missing values.
• The minimal missing values in Age indicate that this column needs imputation during the
data cleaning process.
• Overall, the dataset is relatively complete and ready for preprocessing with minor
adjustments.

19031752 | Sushil Dhital | 25

FC7P01NI MSc Project

Step 5: Validating Data Types

To verify the data types of each column to ensure compatibility with analysis methods and identify
any columns requiring transformation.

Figure 11: Data Types of the Dataset

Observations:

• The column Gender, Location, Blood Group, and Blood Pressure are of object type,
indicating they are categorical or textual data.
• The Age column is of float64 type, which is appropriate for numerical analysis.
• The Total Donations column is of int64 type, suitable for aggregation and statistical
analysis.
• The data types are well-aligned with the dataset’s structure and requirements for further
analysis. Minor preprocessing may be required for categorical columns and text-based
Blood Pressure.

19031752 | Sushil Dhital | 26

FC7P01NI MSc Project

4.2.2 Data Cleaning and Preprocessing

This section focuses on ensuring the dataset is clean and reliable for analysis.

Step 1: Handling Missing Values

This step involves identifying and addressing missing values in the dataset. The Age column,
which had three missing values, was imputed using the median age to maintain data consistency.

Figure 12: Missing Values After Imputation

Observations:

• Missing values in the Age column have been successfully imputed using the median age.
• The dataset now contains no missing values in any column.
• The imputation ensures data consistency and avoids potential biases caused by missing
entries.
• The dataset is now ready for further preprocessing steps, such as handling categorical and
text-based columns.

19031752 | Sushil Dhital | 27

FC7P01NI MSc Project

Step 2: Detecting and Handling Duplicates

This step involves identifying duplicate rows in the dataset and removing them to ensure the data
is free from redundancy. This improves the quality and reliability of the dataset.

Figure 13: Detecting and Handling Duplicates

Observations:

• The dataset contained 250 duplicate rows, which were successfully detected and removed.
• Duplicate rows could have led to biased or inaccurate results if not handled.
• After removing duplicates, the dataset is now free from redundancy, ensuring clean and
consistent data for further analysis.
• This step significantly improves the quality and reliability of the dataset.

19031752 | Sushil Dhital | 28

FC7P01NI MSc Project

Step 3: Validating Categorical Data

This step involves validating the categorical columns (Gender, Location, and Blood Group) to
check for unique values, identify inconsistencies, and ensure data quality.

Figure 14: Validating Categorical Data

Observations:

• The Gender column has two valid unique values: 'Male' and 'Female'.
• The Location column includes diverse geographical regions, such as Kathmandu,
Bhaktapur, Lalitpur, Chitwan, and Makwanpur, among others.
• The Blood Group column contains valid blood group types, including 'A+', 'B+', 'O+',
'AB+', 'O-', and others.
• No inconsistencies or unexpected values were observed in the categorical data, indicating
the data is clean and ready for analysis.

19031752 | Sushil Dhital | 29

FC7P01NI MSc Project

Step 4: Validating Numerical Data

This step involves validating numerical columns (Age and Total Donations) to ensure data
consistency and identify invalid values such as negatives or zeros where they are not expected.

Observations:

• Age Column:
- The mean age is
approximately 37
years, with a range of
18 to 69 years.
- No invalid values
(e.g., negative or
zero) were found in
the Age column.
• Total Donations Column:
- The mean number of
donations is
approximately 7.58,
with a maximum of
104 donations.
- No invalid values
were found, and all
entries are consistent
with the expected
data range.
• Both numerical columns
(Age and Total
Donations) are clean and
Figure 15: Validating Numerical Data
ready for further analysis.

19031752 | Sushil Dhital | 30

FC7P01NI MSc Project

4.2.3 Outlier Detection and Handling

Outliers can skew statistical analyses and affect the accuracy of predictive models. This section
focuses on identifying and addressing outliers in numerical columns, specifically Age and Total
Number of Donation.

Step 1: Visualizing Outliers Using Boxplots

This step includes using boxplots to visually

identify potential outliers in the Age and Total
Donations columns.

Observations:

• Age Column:
- The boxplot for Age shows no
significant outliers.
- The data distribution appears
within the acceptable range of 18
to 69 years.
• Total Donations Column:
- The boxplot for Total Donations
indicates the presence of multiple
outliers, especially values greater
than 40 donations.
- These outliers represent a small
fraction of donors who have
donated at an exceptionally high
frequency.
• The visualizations confirm the need to
address outliers in the Total Donations
column while leaving the Age column
unmodified.
Figure 16: Boxplots for Age and Total Donations

19031752 | Sushil Dhital | 31

FC7P01NI MSc Project

Step 2: Handling Outliers Using the IQR Method

This step includes detecting and handling outliers in the Total Donations column using the
Interquartile Range (IQR) method. The IQR method identifies outliers as values outside the range
Q1−1.5∗IQR, Q3+1.5∗IQR, where Q1 and Q3 are the first and third quartiles, respectively.

Figure 17: Dataset Size Before and After Outlier Removal

Observations:

• The dataset originally had 4,873 records before outlier removal.

• After applying the IQR method to handle outliers in the Total Donations column, the
dataset size was reduced to 4,612 records.
• A total of 261 records were identified as outliers and removed.
• Outlier handling ensures that the dataset is now more representative and free of extreme
values, which could skew the analysis.

19031752 | Sushil Dhital | 32

FC7P01NI MSc Project

4.2.4 Saving the Cleaned Dataset

This section includes saving the cleaned and pre-processed dataset to a new CSV file for future
use. This ensures that the clean data is preserved and can be directly loaded for subsequent analysis.

Step 1: Displaying Cleaned Dataset Information

This step includes displaying the dataset's information after cleaning and preprocessing to confirm
the structure and data types.

Figure 18: Dataset Information After Cleaning

Observations:

• The cleaned dataset contains 4,612 entries across 6 columns: Gender, Age, Location, Blood
Group, Blood Pressure, and Total Donations.
• All columns have non-null values, confirming the absence of missing data after cleaning.
• Data types remain consistent.
• The dataset is clean and ready for further analysis and saving.

19031752 | Sushil Dhital | 33

FC7P01NI MSc Project

Step 2: Displaying Dataset Preview

This step includes displaying the first few rows of the cleaned dataset to visually confirm the
structure and content.

Figure 19: Displaying First Five Rows of the Cleaned Dataset

Observations:

• The first five rows of the cleaned dataset confirm the structure and content of the data:
• The dataset contains clean, consistent, and meaningful values after preprocessing.
• The Age column shows realistic values, ranging from 27 to 48 in the previewed rows.
• The Total Donations column exhibits varying donation frequencies, with values such as 6,
20, and 13 in the preview.
• Categorical columns (Gender, Location, Blood Group) display valid and expected entries.

19031752 | Sushil Dhital | 34

FC7P01NI MSc Project

Step 3: Saving the Cleaned Dataset

This step includes saving the cleaned dataset to a CSV file to preserve the pre-processed data for
further analysis.

Figure 20: Saving the Cleaned Dataset

Figure 21: Screenshot showing saved file of clean dataset

Observations:

• The cleaned dataset has been successfully saved as a CSV file named
cleaned_blood_donation_data.csv.
• The saved file ensures that the pre-processed data is preserved for future analysis and
modelling tasks.
• This step marks the completion of the data cleaning and preprocessing process, providing
a reliable dataset for subsequent steps in the study.

19031752 | Sushil Dhital | 35

FC7P01NI MSc Project

4.3 Data Analysis Techniques

This chapter focuses on the specific techniques and methodologies employed to analyse the
cleaned dataset. The goal is to extract meaningful insights, identify patterns, and address the
research questions and objectives outlined in earlier chapters.

• Exploratory Data Analysis (EDA)

• Predictive Analysis using Supervised Learning
• Statistical Analysis
• Insights and Summary

4.3.1 Exploratory Data Analysis (EDA)

Exploratory Data Analysis (EDA) is a crucial initial step in any data science project. It involves
thoroughly examining datasets to uncover their key characteristics, identify patterns, and explore
relationships between variables. This process helps provide a deeper understanding of the data by
using visualizations and analytical techniques to highlight trends, detect outliers, and reveal
meaningful insights that might otherwise go unnoticed. EDA serves as the foundation for making
informed decisions and building effective data-driven solutions. (nikhilaggarwal3, 2024)

[Link] Age, Gender, and Location Distributions

This topic involves analysing and visualizing the distributions of Age, Gender, and Location in the
dataset to gain insights into the demographic characteristics of blood donors.

Step 1: Histogram for Age Distribution

A histogram is a graphical representation made up of adjacent rectangular bars, where each bar
represents a specific range of data values. The width of each bar corresponds to a class interval,
while the height reflects the frequency of data points within that interval. The horizontal axis
displays the numerical ranges, and the vertical axis represents the frequency, showing how much
data falls into each range. The data used determines the number and size of the intervals. These
shapes help in understanding the data’s overall distribution and patterns. (Cuemath, 2024)

19031752 | Sushil Dhital | 36

FC7P01NI MSc Project

Figure 22: Age Distribution of Donors

Observations:

• The Age distribution shows that most donors are in their 20s and 30s, with the highest
frequency around the 20–30 age group.
• The frequency decreases gradually with age, indicating fewer donors in the older age
groups.
• The histogram is well-distributed, with no unusual spikes or gaps, showing a natural
decline in participation as age increases.

19031752 | Sushil Dhital | 37

FC7P01NI MSc Project

Step 2: Bar Plot for Gender Distribution

Bar charts are versatile tools that can serve different purposes depending on how they are used.
One common use is in data visualization, where they represent categorical data through bars of
varying heights or lengths. The size of each bar corresponds to the value it represents, making it
easy to compare different categories at a glance. (CFI Team, 2024)

Figure 23: Gender Distribution of Donors

Observations:

• The Gender distribution shows a higher number of male donors compared to female
donors.
• Male donors account for approximately two-thirds of the donor population, while females
contribute about one-third.
• This highlights a gender imbalance in donor participation, indicating a need to focus on
increasing female donor engagement.

19031752 | Sushil Dhital | 38

FC7P01NI MSc Project

Step 3: Bar Plot for Location Distribution

This step includes using a bar plot to visualize the distribution of donors across different locations,
providing insights into geographic trends.

Figure 24: Location Distribution of Donors

Observations:

• The majority of donors are from Kathmandu, which contributes the highest frequency,
exceeding 2,000 donors.
• Lalitpur and Bhaktapur follow as the next most significant donor regions, indicating strong
donor activity within the Kathmandu Valley.

19031752 | Sushil Dhital | 39

FC7P01NI MSc Project

[Link] Blood Group Trends and Insights

This topic involves analysing the distribution and trends of blood groups among donors.
Visualizations such as bar plots will be used to highlight key insights.

Step 1: Bar Plot for Blood Group Distribution

Figure 25: Bar Plot for Blood Group Distribution

Observations:

• The most common blood group among donors is O+, with the highest frequency exceeding
1,750 donors.
• Blood groups A+ and B+ are the second and third most prevalent, with significant
representation in the dataset.
• Rare blood groups such as AB-, A-, B-, and O- have the lowest donor frequencies,
highlighting potential challenges in managing rare blood group inventories.

19031752 | Sushil Dhital | 40

FC7P01NI MSc Project

Step 2: Analysing Trends in Blood Groups

Figure 26: Percentage Distribution of Blood Groups

Observations:

• The above pie chart clearly illustrates the dominance of positive blood groups, reflecting
typical population trends and highlighting potential challenges for rare blood group
availability.
• The visual representation effectively communicates the relative proportions of each blood
group.

19031752 | Sushil Dhital | 41

FC7P01NI MSc Project

4.3.2 Statistical and Hypothesis Testing

A basic statistical technique used in data science, machine learning, statistics, and other domains
to make defensible conclusions based on empirical data is hypothesis testing. Using sample
statistics, it entails generating hypotheses on population parameters and carefully comparing them
to the data gathered. (deepanshu_jain, 2024)

[Link] Correlation Matrix Analysis

A correlation matrix is a statistical tool used to evaluate the relationship between two variables
within a dataset. It is presented as a table where each cell contains the correlation coefficient for a
pair of variables. A coefficient of 1 signifies a strong positive relationship, 0 indicates no
relationship, and -1 represents a strong negative relationship. One of the most common
applications of a correlation matrix is in constructing regression models, where it helps identify
how variables are related to one another. (Whitfield, 2024)

Step 1: Creating a Correlation Matrix

Figure 27: Correlation Matrix of Numerical Variables

Observations:

• The correlation matrix indicates a weak positive correlation (0.14) between Age and Total
Donations.
• This suggests that donor age has a minimal influence on the number of donations,
indicating that other factors might play a more significant role in determining donation
frequency.

19031752 | Sushil Dhital | 42

FC7P01NI MSc Project

Step 2: Visualizing Correlation Matrix (Heatmap)

Figure 28: Heatmap of Correlation Matrix

Observations:

• The heatmap confirms the weak positive correlation (0.14) between Age and Total
Donations.
• The diagonal values of 1.0 represent perfect correlations of variables with themselves, as
expected.
• The color gradient (from red to blue) effectively highlights the relative strengths of
correlations, with darker colors showing weaker relationships.
• No strong correlations are observed between the variables, indicating a lack of significant
linear relationships within the dataset's numerical columns.

19031752 | Sushil Dhital | 43

FC7P01NI MSc Project

[Link] T-Test for Gender Based Differences

The t-test is a statistical tool used to assess whether the means of two groups are significantly
different from each other. This method is especially effective when the population variance is
unknown, and the data follows a normal distribution. It is commonly employed in hypothesis
testing to determine if the observed difference between the two group means is genuine or merely
due to random chance. By examining the data, the t-test helps establish whether the differences are
statistically significant and meaningful. (prakharr0y, 2024)

Step 1: Formulating Hypothesis

This step involves performing a T-Test to compare the average number of donations between male
and female donors, testing whether there is a significant gender-based difference. Defining the null
and alternative hypotheses for the T-Test:

Hypotheses:

• Null Hypothesis (H₀): There is no significant difference in the average number of donations
between male and female donors.
• Alternative Hypothesis (H₁): There is a significant difference in the average number of
donations between male and female donors.

Step 2: Performing T-Test

Figure 29: Code to perform T-Test

19031752 | Sushil Dhital | 44

FC7P01NI MSc Project

Figure 30: T-Test Results for Gender-Based Differences in Donations

Observations:

• The T-Statistic is 29.9366, indicating the magnitude of difference between the mean
donations of male and female donors.
• The P-Value is 0.0000, which is below the significance threshold of 0.05.
• The null hypothesis (H₀) is rejected, indicating a statistically significant difference in the
average number of donations between male and female donors.
• Male donors are likely contributing more frequently than female donors, as inferred from
the significant results.

[Link] ANOVA for Blood Group Comparisons

Analysis of Variance (ANOVA) is a statistical technique used to assess whether there are
significant differences between the means of two or more groups. It works by comparing the
variability within each group to the variability between groups. This analysis helps determine if
the differences observed in group means are statistically significant or if they could have occurred
by chance. (gajawada, 2019)

Step 1: Formulating Hypothesis

Defining the null and alternative hypotheses for the ANOVA test:

Hypotheses:

• Null Hypothesis (H₀): The mean number of donations is the same across all blood groups.
• Alternative Hypothesis (H₁): At least one blood group has a significantly different mean
number of donations.

19031752 | Sushil Dhital | 45

FC7P01NI MSc Project

Step 2: Performing ANOVA Test

Figure 31: Code to perform ANOVA Test

Figure 32: ANOVA Test Results for Blood Group Comparisons

Observations:

• The F-Statistic is 7.8629, indicating variability in the mean number of donations among
blood groups.
• The P-Value is 0.0000, which is below the significance threshold of 0.05.
• The null hypothesis (H₀) is rejected, indicating that at least one blood group has a
significantly different mean number of donations.
• These findings highlight differences in donor behaviour across blood groups, suggesting
the need for further investigation to identify specific groups contributing to the variation.

19031752 | Sushil Dhital | 46

FC7P01NI MSc Project

4.3.3 Predictive Modelling

Predictive analytics is a field of data science that uses statistical techniques, machine learning
algorithms, and historical data to produce data-driven predictions about the future. (ainykhan1410,
2024)

[Link] Preparing the Dataset

This step ensures the data is prepared for predictive modelling by providing separate datasets for
training the model and evaluating its performance.

Step 1: Splitting Data into Training and Testing Sets

Figure 33: Training and Testing Dataset Split Overview

Observations:

• The dataset was successfully encoded, with categorical columns Gender, Location, and
Blood Group converted into numerical values.
• The dataset was split into training and testing sets:
- Training Set: 3,689 rows
- Testing Set: 923 rows

19031752 | Sushil Dhital | 47

FC7P01NI MSc Project

[Link] Linear Regression

This topic involves applying linear regression to predict donor behaviour based on numerical
variables. It includes data preparation, training the model, and evaluating its performance.

Step 1: Training the model

Figure 34: Linear Regression Model Coefficients

Observations:

• The model was successfully trained using the training dataset.

• The intercept of the regression model is approximately 0.576, indicating the baseline
prediction when all features are zero.
• The coefficients for the features are as follows:
- Age: 0.0738 – Indicates a positive relationship with Total Donations.
- Gender: 4.1014 – Suggests males (encoded as 1) tend to donate more frequently than
females (encoded as 0).
- Location: -0.0918 – Indicates a slight negative influence of location on donations.
- Blood Group: 0.0945 – Suggests a small positive relationship between blood group and
donation frequency.

19031752 | Sushil Dhital | 48

FC7P01NI MSc Project

Step 2: Evaluating the Model

Figure 35: Evaluation Metrics for Linear Regression Model

Observations:

• Mean Absolute Error (MAE): The MAE is 4.53, indicating the average absolute difference
between the predicted and actual values.
• Mean Squared Error (MSE): The MSE remains 33.01, reflecting the squared error
magnitude.
• R-squared (R²): The R² value is 0.13, meaning the model explains 13% of the variance in
Total Donations.
• The relatively low R² score and moderate MAE suggest that the Linear Regression model
is limited in its predictive power, potentially due to insufficient features or linear
assumptions.

19031752 | Sushil Dhital | 49

FC7P01NI MSc Project

[Link] Random Forest

Random Forest Regression is a versatile machine learning technique used for predicting numerical
values. It works by combining the predictions of multiple decision trees, which helps to enhance
accuracy and reduce the risk of overfitting. This ensemble approach ensures more reliable and
robust predictions. (avik_dutta, 2024)

Step 1: Training the Model

Figure 36: Training of Random Forest Model

Observations:

• The Random Forest model was successfully trained using the provided training dataset.
• Using 100 estimators ensures the model captures complex patterns in the data, improving
predictive performance.

19031752 | Sushil Dhital | 50

FC7P01NI MSc Project

Step 2: Evaluating the Model

Figure 37: Evaluation Metrics for Random Forest Model

Observations:

• Mean Absolute Error (MAE): The Random Forest model achieved an MAE of 4.97,
indicating the average error magnitude between predictions and actual values.
• Mean Squared Error (MSE): The MSE is 41.37, reflecting the squared error magnitude.
• R-squared (R²): The R² value is -0.09, which is negative, indicating that the model performs
worse than a simple mean-based prediction.
• The results suggest that the Random Forest model is not effective for this dataset, possibly
due to overfitting or insufficient feature relevance.

19031752 | Sushil Dhital | 51

FC7P01NI MSc Project

[Link] Gradient Boosting

Using gradient descent, each new model is trained to minimise the loss function, such as mean
squared error or cross-entropy of the prior model. Gradient Boosting is a potent boosting method
that turns several weak learners into strong learners. (nikki2398, 2023)

Step 1: Training the Model

Figure 38: Training of Gradient Boosting Model

Observations:

• The Gradient Boosting model was successfully trained using the training dataset.
• The model is configured with the following parameters:
- n_estimators: 100 (Number of boosting stages).
- learning_rate: 0.1 (Controls the contribution of each tree).
• Gradient Boosting uses a combination of decision trees to reduce errors iteratively, making
it robust for predictive modeling.

19031752 | Sushil Dhital | 52

FC7P01NI MSc Project

Step 2: Evaluating the Model

Figure 39: Evaluation Metrics for Gradient Boosting Model

Observations:

• Mean Absolute Error (MAE): The Gradient Boosting model achieved an MAE of 4.50,
indicating a lower average error magnitude compared to previous models.
• Mean Squared Error (MSE): The MSE is 32.84, slightly better than the Linear Regression
and Random Forest models.
• R-squared (R²): The R² value is 0.13, meaning the model explains 13% of the variance in
Total Donations.
• While the performance of Gradient Boosting is comparable to Linear Regression, the MSE
is slightly reduced, indicating a marginal improvement in prediction accuracy.

19031752 | Sushil Dhital | 53

FC7P01NI MSc Project

[Link] Support Vector Regression (SVR)

Support Vector Machines (SVM) are frequently used in machine learning to solve classification
issues, but Support Vector Regression (SVR) allows them to be used to solve regression problems
as well. SVR is based on the same concepts as SVM, but instead of categorizing data points, it
focusses on predicting continuous outputs. (Sethi, 2024)

Step 1: Training the Model

Figure 40: Training of Support Vector Regression Model

Observations:

• The Support Vector Regression (SVR) model was successfully trained using the training
dataset.
• Key parameters for the model:
- Kernel: rbf (Radial Basis Function), suitable for capturing non-linear relationships.
- C: 1.0 (Regularization parameter), balancing the trade-off between achieving a low
error and minimizing model complexity.
- Epsilon: 0.1, specifying the margin within which no penalty is given for errors.

19031752 | Sushil Dhital | 54

FC7P01NI MSc Project

Step 2: Evaluating the Model

Figure 41: Evaluation Metrics for Support Vector Regression Model

Observations:

• Mean Absolute Error (MAE): The SVR model achieved an MAE of 4.61, indicating the
average error magnitude.
• Mean Squared Error (MSE): The MSE is 42.47, reflecting the squared error magnitude.
• R-squared (R²): The R² value is -0.12, which is negative, indicating that the model performs
worse than a simple mean-based prediction.
• The performance metrics suggest that SVR may not be well-suited for this dataset,
potentially due to insufficient feature relevance or non-linearity in the data.

19031752 | Sushil Dhital | 55

FC7P01NI MSc Project

[Link] XGBoost

XGBoost is a powerful machine learning algorithm that falls under the gradient boosting
framework, a type of ensemble learning method. It leverages decision trees as base models and
uses regularization techniques to improve its ability to generalize and avoid overfitting. Renowned
for its computational efficiency, XGBoost is capable of handling missing data seamlessly,
providing valuable insights through feature importance analysis, and processing large datasets
effectively. These attributes make it particularly well-suited for tackling complex predictive
challenges. (Tyagi, 2024)

Step 1: Parameter Optimization and Training

Figure 42: Hyperparameter Tuning and Training of XGBoost Model

19031752 | Sushil Dhital | 56

FC7P01NI MSc Project

Observations:

• The GridSearchCV was successfully performed over 8 candidate combinations of

hyperparameters using 5-fold cross-validation.
• The best hyperparameters identified are:
- learning_rate: 0.01
- max_depth: 3
- n_estimators: 200

Step 2: Evaluating the Model

Figure 43: Evaluation Metrics for XGBoost Model

Observations:

• Mean Absolute Error (MAE): The XGBoost model achieved an MAE of 4.50, which is
comparable to Gradient Boosting.
• Mean Squared Error (MSE): The MSE is 32.75, slightly better than Gradient Boosting and
other models evaluated.

19031752 | Sushil Dhital | 57

FC7P01NI MSc Project

• R-squared (R²): The R² value is 0.14, indicating that the model explains 14% of the
variance in Total Donations. This is the highest R² among the models evaluated so far.

4.3.4 Clustering Analysis

Since cluster analysis is an unsupervised learning process, you are unaware of the number of
clusters in the data until the model is run. Cluster analysis, in contrast to many other statistical
techniques, is usually applied in situations where no assumptions are made on the potential
relationships present in the data. It gives details on the locations of associations and patterns in the
data, but it doesn't explain what they could be or what they signify. (Webster, 2025)

[Link] Data Preparation for Clustering

Figure 44: Data Preparation Summary for Clustering Analysis

Observations:

• Relevant features, Age and Total Donations, were selected for clustering analysis.
• The features were standardized using StandardScaler to ensure they have a mean of
approximately 0 and a standard deviation of 1.
• Standardization ensures that both features contribute equally to the clustering process,
avoiding bias due to scale differences.

19031752 | Sushil Dhital | 58

FC7P01NI MSc Project

[Link] Determining Optimal Number of Clusters

Figure 45: Code to Determine Optimal Number of Clusters

Figure 46: Elbow Curve for Determining Optimal Number of Clusters

19031752 | Sushil Dhital | 59

FC7P01NI MSc Project

Observations:

• The Elbow Curve shows the inertia (within-cluster sum of squares) plotted against the
number of clusters (K).
• There is a noticeable "elbow" point at K = 4, where the inertia decreases significantly
before flattening out.
• Based on the Elbow Method, the optimal number of clusters for this dataset is 4.

[Link] Applying K-Means Clustering

Figure 47: Clustering Results with Assigned Cluster Labels

Observations:

• The K-Means algorithm successfully segmented the donors into 4 clusters as identified by
the Elbow Method.
• Cluster labels were assigned to each donor, enabling segmentation based on Age and Total
Donations.
• The clustering results provide insights into group characteristics that can be used for
targeted campaigns or resource allocation.

19031752 | Sushil Dhital | 60

FC7P01NI MSc Project

[Link] Visualizing Clustering Results

K-Means Clustering is an unsupervised learning technique that organizes unlabeled data into
distinct groups or clusters. The parameter K specifies the number of clusters to create; for example,
setting K=2 results in two clusters, while K=3 forms three clusters, and so forth. This method uses
an iterative process to group data points based on shared characteristics, ensuring that each point
belongs to one cluster. K-Means is a practical and efficient tool for identifying patterns in data, as
it automatically groups similar data points without requiring labeled datasets or prior training,
making it highly effective for exploring and understanding large, unlabelled datasets. (Javatpoint,
2024)

Figure 48: Code to Visualize for Age vs Total Donations

19031752 | Sushil Dhital | 61

FC7P01NI MSc Project

Figure 49: K-Means Clustering Visualization for Age vs Total Donations

Observations:

• The scatter plot provides a clear visualization of the 4 clusters segmented by the K-Means
algorithm.
• Each cluster represents a distinct group of donors based on their Age and Total Donations.
• The clusters are color-coded, enabling easy interpretation of donor group characteristics:
- Cluster 0: Donors with moderate age and relatively higher donation frequencies.
- Cluster 1: Younger donors with fewer donation frequencies.
- Cluster 2: Older donors with moderate donation frequencies.
- Cluster 3: Donors across varying ages with low donation frequencies.
• These clusters can be leveraged for tailored donor engagement and resource allocation.

19031752 | Sushil Dhital | 62

FC7P01NI MSc Project

Chapter 5: System Prototype

5.1 Survey Justifying the Need for the Application

To evaluate the need for a mobile application addressing challenges in Nepal’s blood donation
management systems, a survey titled "Improving Blood Donation Management: Feedback from
Doctors" was conducted using Google Forms. The survey aimed to gather insights from licensed
medical practitioners on the current difficulties in sourcing blood during emergencies and the
potential value of proposed app features. This step was critical in aligning the app design with
practical requirements and user expectations.

Figure 50: Google Form Survey

• Survey Objective

The primary objective of the survey was to assess whether the proposed mobile application,
BloodConnect, could address the key challenges faced by doctors in sourcing blood. These
challenges included donor availability, communication gaps, and the lack of real-time
coordination during emergencies.

19031752 | Sushil Dhital | 63

FC7P01NI MSc Project

• Survey Design

The survey was designed to focus on major pain points in the blood donation process and to
gather feedback on proposed features such as locating blood banks, searching for blood by
group, and accessing emergency contacts. Questions were structured to identify priorities and
expectations for a mobile application dedicated to blood donation management.

• Participants

The survey targeted doctors licensed under the National Medical Council (NMC) of Nepal.
Their professional experience provided valuable insights into the practical challenges of
sourcing and managing blood donations during emergencies.

5.2 Survey Results and Responses

This section presents the findings from the survey conducted with doctors to evaluate the necessity
and relevance of the BloodConnect mobile application. The responses collected through structured
questions provide valuable insights into the challenges faced by doctors in blood donation
management and their feedback on the proposed app features. Each question and its corresponding
responses are detailed below.

• Question 1: What is your Name?

The first question collected the names of

the participating doctors. A total of 23
responses were recorded, ensuring that the
feedback was sourced from a diverse
group of medical professionals. This
helped validate the authenticity and
relevance of the survey.

Figure 51: Survey Question 1

19031752 | Sushil Dhital | 64

FC7P01NI MSc Project

• Question 2: Have you faced difficulties in finding blood donors or required blood during
emergencies?

Figure 52: Survey Question 2

The response revealed that 82.6% of the respondents answered ‘Yes’ highlighting the
significant challenges doctors face in sourcing blood during emergencies. A smaller percentage
responded ‘No’ or indicated occasional difficulties underscoring the critical need for a
streamlined solution like the proposed app.

• Question 3: Have you faced difficulties in finding blood donors or required blood during
emergencies?

Figure 53: Survey Question 3

The response revealed that 39.1% of respondents stated, ‘Always’ while 34.8% indicated
‘Often’. This highlights the frequent reliance on blood banks during emergencies. A smaller
percentage chose ‘Occasionally’ or ‘Rarely’ emphasizing the critical role of blood banks in
managing emergency situations.

19031752 | Sushil Dhital | 65

FC7P01NI MSc Project

• Question 4: What is the most significant challenge you face in sourcing blood during
emergencies?

Figure 54: Survey Question 4

The response revealed that 65.2% identified the lack of immediate access to donor or blood
bank information as the biggest challenge. 47.8% mentioned difficulty in finding specific
blood groups, while 17.4% and 8.7% highlighted delayed response times and inefficient
communication, respectively.

• Question 5: Do you think a mobile app with centralized blood bank information would
address current challenges?

Figure 55: Survey Question 5

The response revealed that 73.9% agreed that a mobile app with centralized blood bank
information would address current challenges, while 26.1% expressed uncertainty.

19031752 | Sushil Dhital | 66

FC7P01NI MSc Project

• Question 6: How useful would a feature to send blood donation requests directly from an
app be?

Figure 56: Survey Question 6

The response revealed that 52.2% found the feature to send blood donation requests directly
from an app useful, while 47.8% considered it very useful.

• Question 7: How helpful would it be to view blood donation requests posted by others?

Figure 57: Survey Question 7

The response revealed that 60.9% found viewing blood donation requests posted by others very
helpful, while 30.4% considered it helpful.

19031752 | Sushil Dhital | 67

FC7P01NI MSc Project

• Question 8: Would a feature to search for blood availability by blood groups improve
your practice?

Figure 58: Survey Question 8

The response revealed that 78.3% believed that a feature to search for blood availability by
blood groups would improve their practice, while 17.4% were unsure.

• Question 9: How important is the availability of emergency contact information during

crises?

Figure 59: Survey Question 9

The response revealed that 95.7% considered the availability of emergency contact information
during crises very important, with a small percentage marking it as important.

19031752 | Sushil Dhital | 68

FC7P01NI MSc Project

• Question 10: Would a feature that shows nearby blood banks on a map be beneficial?

Figure 60: Survey Question 10

The response revealed that 78.3% found a feature showing nearby blood banks on a map very
beneficial, while 21.7% considered it beneficial.

• Question 11: How likely are you to use an app that provides detailed information about
nearby blood banks (e.g., contact, inventory)?

Figure 61: Survey Question 11

The response revealed that 65.2% indicated they were very likely to use an app providing
detailed information about nearby blood banks, while 30.4% said they were likely to use it.

19031752 | Sushil Dhital | 69

FC7P01NI MSc Project

5.3 Overall Survey Results and Insights

This section summarizes the key findings from the survey, highlighting the overall necessity and
relevance of the BloodConnect mobile application as perceived by the participating doctors. The
insights derived from the survey responses validate the app's potential impact on addressing
challenges in blood donation management and emergency response systems.

5.3.1 Key Challenges Identified

The survey highlighted several key challenges faced by doctors in blood donation management,
with the most significant being the lack of immediate access to donor and blood bank information,
as noted by 65.2% of participants. Many respondents also emphasized the difficulty in locating
specific blood groups during emergencies which often led to delayed response times. These
challenges underline the pressing need for a streamlined system that provides real-time access to
critical resources, enabling faster and more efficient coordination between donors, blood banks,
and healthcare providers.

5.3.2 Validation of Proposed Features

The survey results strongly validated the relevance of the proposed features for the BloodConnect
mobile application. A majority of respondents (73.9%) expressed confidence that a centralized
mobile application could effectively address the existing challenges in blood donation
management. Features such as the ability to search for blood by group and view nearby blood
banks on a map were deemed highly beneficial by 78.3% of participants. Additionally, an
overwhelming 95.7% highlighted the importance of providing emergency contact information
directly within the app. The capability to send and receive blood donation requests through the app
was also widely appreciated, underscoring the potential of such a platform to enhance
communication, donor engagement, and emergency response efficiency.

5.3.3 Likelihood of Usage

The survey revealed a strong interest and high potential for adoption of the BloodConnect mobile
application among participants. An impressive 95.6% of respondents indicated that they were
likely or very likely to use an app that provides detailed information about blood banks. This

19031752 | Sushil Dhital | 70

FC7P01NI MSc Project

overwhelming response highlights the demand for a digital solution to streamline blood donation
processes and improve accessibility to vital resources, reinforcing the app’s relevance and
practicality.

5.4 Mobile Application (iOS) Name and Logo

The mobile application developed as part of this study is named BloodConnect. The name
symbolizes the app's core purpose of bridging the gap between blood donors, medical staff, and
healthcare providers, thereby enhancing blood donation management and emergency response
systems.

Figure 62: iOS App Logo

The logo of BloodConnect is a visual representation of its mission: "Connect, Donate & Save
Lives." It features:

- A blood pressure monitor - inspired droplet at one end, symbolizing healthcare and
blood donation.
- A heartbeat line transitioning into a heart with a droplet, representing the life-saving
impact of blood donation.
- The tagline "Connect, Donate & Save Lives", which captures the essence of the app's
purpose.

This thoughtfully crafted logo not only symbolizes the connection between donors and recipients
but also emphasizes the importance of saving lives through blood donation

19031752 | Sushil Dhital | 71

FC7P01NI MSc Project

5.5 Design and Features

5.5.1 Development Tools and Approach

The development of BloodConnect was carried out using Swift as the programming language and
Xcode as the integrated development environment (IDE). The design followed an incremental
development approach, which allowed for the gradual addition of features and ensured consistent
testing and refinement at every stage. This methodology was selected for its flexibility and ability
to incorporate user feedback during the design process.

Key considerations in the development process included:

• User-Centric Design: Focusing on intuitive navigation and ease of use for all stakeholders
(Staff, Donors, Doctors).
• Responsive Interface: Ensuring compatibility with varying screen sizes of iOS devices.
• Scalability: Designing the interface to accommodate future backend and API integrations.

5.5.2 Core Features of the App (Design)

The BloodConnect app has distinct features for each user group (Staff, Donors, and Doctors),
ensuring tailored functionalities.

Features Staff Donors Doctors

Login & Registration Secure login with PIN Secure login with PIN Secure login with PIN
along with recovery along with recovery along with recovery
PIN feature. PIN feature. PIN feature.
Dashboard • View blood • Send donation • Search blood by
shortage areas requests group
• Send alerts • View alerts • Contact nearby
• Notify donors • Where to Donate blood banks
• Manage pending list
donation requests
Map Integration Display nearby blood Display nearby blood Display nearby blood
banks banks banks

19031752 | Sushil Dhital | 72

FC7P01NI MSc Project

Alerts Urgent blood Urgent blood Urgent blood

requirements in requirements in requirements in
hospitals hospitals hospitals
Profile Management • Edit profile • Edit profile • Edit profile
• Change PIN • Change PIN • Change PIN
• Emergency • Emergency • Emergency
contacts contacts contacts
• Feedback • Donation History • Feedback
• Accepted Donors • Feedback • My Appointments
• Logout • Donate Fund • Logout
Table 3: Core Features of BloodConnect for Different Users

The BloodConnect app begins with a secure Login and Registration system. Users authenticate
using a PIN, with the option to recover their PIN via the ‘Forgot your PIN’ feature if needed. Each
user type (Staff, Donor, Doctor) is presented with a tailored Dashboard, ensuring access to relevant
features. For instance, staff can manage blood shortages, notify donors, and review pending
requests, while donors can send donation requests and view donation opportunities. Doctors can
search for blood groups and contact nearby blood banks.

An interactive Map Integration helps all users locate nearby blood banks, while the Alerts page
lists urgent blood requirements across hospitals. Additionally, the Profile Management section
allows users to edit their details, update login credentials, manage emergency contacts, provide
feedback, and logout. These features collectively ensure a streamlined and user-friendly
experience for all stakeholders.

19031752 | Sushil Dhital | 73

FC7P01NI MSc Project

5.6 Screenshots of User Interface

This section highlights the visual design of the BloodConnect app through screenshots, showcasing
its user-friendly interface tailored for each user.

• Login and Registration Pages

Figure 63: Login and Registration Pages

The Login and Registration Pages of the BloodConnect app feature a user-friendly design for
seamless onboarding. The login page allows users to securely access their accounts using a 4-digit
PIN, with an option to recover forgotten PINs. The registration pages are designed to serve donors,
doctors, and staff, capturing essential details to ensure a personalized experience and efficient user
management.

19031752 | Sushil Dhital | 74

FC7P01NI MSc Project

• Dashboards

Figure 64: Dashboards of all three users

The Dashboards in the BloodConnect app provide tailored

functionalities for different user groups.

• Donor Dashboard: Displays a ‘Send Blood Donation

Request’ option, guidelines for donation, and a ‘Where
& Where to Donate’ section for nearby blood donation
opportunities.
• Doctor Dashboard: Enables quick blood group
selection and displays a list of nearby blood banks for
direct contact.
• Staff Dashboard: Highlights blood shortage areas,
enables notifying donors, and displays a list of pending
donation requests for management.

19031752 | Sushil Dhital | 75

FC7P01NI MSc Project

• Map Integration

The Map Integration feature in the BloodConnect app provides an

interactive interface for users to locate nearby blood banks and
hospitals. Each location is marked with a pin, and users can click
on a pin to view the name of the facility and get directions directly.
This feature ensures easy navigation and accessibility, making it
convenient for donors, staff, and doctors to find blood banks
efficiently.

• Alert Page

Figure 65: Map Integration

Figure 66: Alert Screens

The Alert Page ensures real-time communication for blood donation emergencies. For both donors
and staff, it lists urgent blood requirements across hospitals, including details like location,
required blood group, and a quick-call option. The doctor-specific alert page displays donation
requests from donors, allowing doctors to directly connect with them for immediate coordination.

19031752 | Sushil Dhital | 76

FC7P01NI MSc Project

• Profile Management

Figure 67: Profile Management Pages of all three users

The Profile Management section in the BloodConnect app offers essential tools for managing user
accounts. Features include the ability to update login PINs, manage emergency contacts, view
donation history, and provide feedback. For doctors, additional options such as viewing
appointments and accepted donors are available. This section ensures users can personalize and
manage their profiles efficiently while also promoting fund donations to support the cause.

19031752 | Sushil Dhital | 77

FC7P01NI MSc Project

• Other Screens

Figure 68: Other Inner Screens

The other pages include Change Pin, Edit Profile, Emergency Numbers, User Feedback, Forgot
Pin and Logout.

19031752 | Sushil Dhital | 78

FC7P01NI MSc Project

5.7 Mobile App Limitations and Future Work

While the BloodConnect mobile application prototype demonstrates a user-friendly design and
thoughtfully tailored features for donors, doctors, and staff, it remains in the conceptual stage. The
current version focuses on the app’s design and layout, with no backend development or real-time
functionality. This section highlights its limitations and outlines potential future enhancements to
transform the prototype into a fully functional system.

5.7.1 Current Limitation

The BloodConnect mobile application prototype, as developed in this thesis, is limited to the
design phase and lacks backend integration, such as API connections and database support, which
are essential for enabling real-time data synchronization and user account functionality. While the
design includes core features like donor registration, blood bank search, and emergency alerts,
these functionalities are conceptual and not implemented in the current version. Additionally,
advanced features such as push notifications, blood request tracking, and analytics remain untested
and undeveloped. Furthermore, the app is restricted to the iOS platform, limiting its accessibility
to a broader audience, particularly in Nepal, where Android devices are widely used. These
limitations emphasize the prototype’s role as a foundational concept, providing a blueprint for
future development.

5.7.2 Future Work

To fully realize the potential of the BloodConnect app, several future developments are proposed
to enhance its functionality and impact. Backend integration, including APIs and a centralized
database, will enable real-time communication and dynamic data management, transforming the
app from a prototype into a functional system. Expanding the app to Android devices will
significantly increase accessibility, ensuring it reaches a wider user base in Nepal. Additionally,
advanced features such as blood inventory tracking, predictive analytics for blood demand, and
push notifications for urgent alerts will further improve efficiency and responsiveness.
Collaboration with blood banks and hospitals will be essential to ensure a seamless flow of
information and provide the necessary support for the app’s successful implementation.

19031752 | Sushil Dhital | 79

FC7P01NI MSc Project

Chapter 6: Results and Findings

6.1 Data Analysis Results

This section presents the findings from the data analysis process. The results are divided into three
main categories: Exploratory Data Analysis (EDA), Predictive Modelling, and Clustering
Analysis. Each subsection highlights key insights gained through statistical analysis,
visualizations, and machine learning techniques.

6.1.1 Exploratory Data Analysis (EDA) Results

The exploratory data analysis provided valuable insights into donor demographics and geographic
patterns. Age emerged as a key factor influencing donation frequency, with younger donors
contributing less frequently compared to their older counterparts. Geographic trends highlighted
disparities in donation rates across regions in the Bagmati area, with some areas showing
significantly lower participation. Blood group distribution followed expected patterns; however,
critically underrepresented groups like AB-negative highlighted the need for targeted
mobilization. These findings lay the groundwork for crafting focused donor engagement strategies
and optimizing resource allocation.

6.1.2 Predictive Modelling Results

This subsection highlights the performance results of various predictive models used to estimate
Total Donations based on donor characteristics. The models evaluated include Linear Regression,
Random Forest, Gradient Boosting, Support Vector Regression (SVR), and XGBoost. Their
performance is assessed using key evaluation metrics, including Mean Absolute Error (MAE),
Mean Squared Error (MSE), and R-squared (R²), providing a comprehensive comparison of their
effectiveness in predicting donation behaviour.

Model MAE MSE R2 Insights

Linear Regression 4.53 33.01 0.13 Limited predictive capability, explains
only 13% of variance.
Random Forest 4.97 41.37 -0.09 Underperformed, likely due to overfitting
or insufficient feature relevance.

19031752 | Sushil Dhital | 80

FC7P01NI MSc Project

Gradient Boosting 4.50 32.84 0.13 Comparable to Linear Regression, slightly

better MSE.
Support Vector 4.61 42.47 -0.12 Poor performance, failed to capture data
Regression (SVR) patterns effectively.
XGBoost 4.50 32.75 0.14 Best performing model, explains 14% of
variance with the lowest MSE among all
tested models.
Table 4: Comparison of Predictive Models

The predictive models employed in this study were Linear Regression, Random Forest, Gradient
Boosting, and XGBoost which provided a deeper understanding of donor behaviour and donation
frequency. Among these, XGBoost stood out for its high accuracy and ability to capture complex
patterns in the data. Analysing feature importance revealed that factors such as donor age and prior
donation history played a significant role in predicting donation frequency. These results validate
the potential of predictive modelling as a powerful tool for making data-driven decisions in blood
donation management.

6.1.3 Clustering Results

K-Means clustering segmented donors into four distinct groups, offering meaningful insights into
their behaviour. For instance, younger donors with low donation frequency formed one cluster,
while older, consistent donors formed another. This segmentation provides a clear framework for
designing tailored campaigns to engage fewer active donors while maintaining strong relationships
with frequent donors. These results underscore the value of clustering in identifying patterns that
can inform effective donor mobilization strategies.

19031752 | Sushil Dhital | 81

FC7P01NI MSc Project

6.2 Mobile Application Prototype Results

This section evaluates the results of the system development process, focusing on the functionality
and design outcomes of the BloodConnect app.

Figure 69: Screenshot of Domain

6.2.1 Survey Insights

As detailed in Chapter 5.2, feedback from licensed doctors, gathered through the survey, strongly
validated the relevance of the proposed mobile application. A significant 73.9% of respondents
expressed confidence in the app’s ability to address challenges in blood donation management.
Features like geolocation-based donor matching, blood bank search, and real-time emergency
notifications were highly rated for their practicality and potential impact. These findings confirmed
that the app design aligns closely with the needs of healthcare professionals and stakeholders.

6.2.2 Prototype Evaluation

As detailed in Chapter 5.6, while the BloodConnect app remains a design prototype, it successfully
demonstrates the potential to address inefficiencies in blood donation management. The user-
friendly interface and tailored features for donors, doctors, and staff were well-received during the
evaluation. However, without backend integration or real-time functionality, the prototype serves
as a conceptual demonstration rather than an operational system. These results highlight the app’s
promise while emphasizing the importance of further development to fully realize its potential.

19031752 | Sushil Dhital | 82

FC7P01NI MSc Project

6.3 Dashboard Implementation

The dashboard developed for this study serves as an interactive tool to visualize and explore the
key findings derived from the research. Built using Python Streamlit, the dashboard provides an
intuitive interface that allows users to engage with the data, uncover insights, and better understand
donor behaviour, geographic disparities, and predictive modelling outcomes.

Designed to be user-friendly and accessible, the dashboard facilitates real-time interaction with the
study’s results. Stakeholders can navigate through sections such as exploratory data analysis
(EDA), predictive modelling results, and clustering insights. Each section is complemented with
visualizations like histograms, bar plots, scatter plots, and comparison tables, making complex
data more comprehensible. By presenting the findings in an engaging and dynamic format, the
dashboard bridges the gap between raw data and actionable insights, ensuring that the study's
conclusions are accessible and practical for decision-making.

Figure 70: Dashboard Implementation

19031752 | Sushil Dhital | 83

FC7P01NI MSc Project

6.3.1 Exploratory Data Analysis Dashboard

Figure 71: Exploratory Data Analysis Dashboard

The Exploratory Data Analysis (EDA) section of the dashboard provides a comprehensive
visualization of key demographic and behavioural patterns among donors. It features a histogram
of age distribution, highlighting the concentration of donors within specific age ranges, which
helps identify demographic trends. A bar plot illustrating gender distribution reveals disparities
between male and female donors, offering insights into engagement gaps. Geographic donation
patterns are presented through a bar plot of location-based participation, showcasing higher donor
density in urban regions like Kathmandu compared to rural areas. Additionally, a visual
representation of blood group distribution highlights the prevalence of common groups such as O+
and the critical scarcity of rare groups like AB- and B-. Finally, a line graph displaying trends in
blood group donations over time offers actionable insights into shifting patterns, further supporting
targeted donor mobilization strategies.

19031752 | Sushil Dhital | 84

FC7P01NI MSc Project

6.3.2 Predictive Modelling Dashboard

Figure 72: Predictive Modelling Dashboard

The Predictive Modeling section of the dashboard presents detailed metrics and visualizations for
various machine learning models, including Linear Regression, Random Forest, Gradient
Boosting, Support Vector Regression (SVR), and XGBoost. Each model’s performance is
displayed through evaluation metrics such as Mean Squared Error (MSE), Mean Absolute Error
(MAE), and R², enabling a comparative analysis of their predictive accuracy. Visualizations of
feature importance or coefficients further illustrate the impact of key variables, such as donor age
and prior donation history, on total donation predictions. XGBoost, identified as the best-
performing model, is highlighted for its superior accuracy and ability to capture complex patterns
within the dataset. This section of the dashboard provides stakeholders with actionable insights
into the predictive capabilities of each model, supporting data-driven decision-making in blood
donation management.

19031752 | Sushil Dhital | 85

FC7P01NI MSc Project

6.3.3 Model Comparison Dashboard

Figure 73: Model Comparison Dashboard

The Model Comparison section of the dashboard offers a detailed evaluation of the predictive
performance of various machine learning models, including Linear Regression, Random Forest,
Gradient Boosting, Support Vector Regression (SVR), and XGBoost. Metrics such as Mean
Absolute Error (MAE), Mean Squared Error (MSE), and R² are presented in a summary table,
making it easy to compare the accuracy and effectiveness of each model. Visual highlights
showcase XGBoost as the top-performing model, achieving the lowest MAE and MSE while
delivering the highest R² value. This demonstrates its capability to handle complex patterns within
the dataset effectively. The comparison not only confirms XGBoost’s suitability for predicting
donation behaviour but also sheds light on the strengths and limitations of other models, providing
valuable insights to guide future analytical strategies.

19031752 | Sushil Dhital | 86

FC7P01NI MSc Project

6.3.4 Clustering Dashboard

Figure 74: Clustering Dashboard

The Clustering section of the dashboard provides an in-depth visualization of donor segmentation
using K-Means clustering. It begins with an Elbow Curve, which identifies the optimal number of
clusters (K = 4) by analysing the variance explained by different cluster numbers. A Cluster
Summary is presented in a tabular format, outlining the key characteristics of each donor group,
such as age, donation frequency, and behavioural trends. Finally, a scatter plot visualizes the
segmentation, showing donor clusters based on age and total donations. This clear and interactive
representation allows stakeholders to understand patterns such as high-value consistent donors or
younger, infrequent donors, offering actionable insights for targeted engagement and resource
optimization.

19031752 | Sushil Dhital | 87

FC7P01NI MSc Project

Chapter 7: Discussion

7.1 Interpretation of Results

The exploratory data analysis (EDA) revealed significant demographic and geographic trends in
blood donation behaviour. Male donors were found to significantly outnumber female donors,
highlighting potential gender-based disparities in donor engagement. Donors were predominantly
within the 20–50 age range, with fewer contributions from younger and older demographics. Blood
group O+ had the highest representation, while rare blood groups like AB- and B- were critically
underrepresented. Geographically, Kathmandu had the highest donor density, whereas rural areas
showed significantly lower participation, emphasizing the need for targeted donor engagement in
underserved regions.

Predictive modelling provided valuable insights for donor behaviour analysis and donation
frequency predictions. Among the tested models, XGBoost demonstrated the best performance,
achieving the lowest Mean Squared Error (MSE) of 32.75 and the highest R² of 0.14, making it
the most suitable for predicting total donations. Other models, such as Random Forest and Support
Vector Regression (SVR), underperformed due to feature limitations or the complexity of the
dataset. These results highlight the potential of advanced machine learning techniques for
improving resource allocation and donor mobilization strategies.

Clustering analysis segmented donors into four meaningful clusters based on age and donation
frequency. These included high-value, consistent donors; younger donors with growth potential;
older, reliable donors; and infrequent donors requiring more engagement. These clusters provide
actionable insights for designing targeted campaigns and optimizing resource allocation by
focusing on specific donor groups. For instance, strategies can be developed to engage younger
donors while maintaining the loyalty of consistent contributors.

The research questions were effectively addressed through these methods. Geographic and
demographic disparities were identified (R1), predictive modelling and clustering provided data-
driven strategies for donor mobilization (R2), and the findings laid the foundation for integrating
these insights into a mobile application prototype (R3). This combination of analytical techniques
and design conceptualization aligns with the study's objectives, offering a comprehensive approach
to addressing the inefficiencies in Nepal’s blood donation management systems.

19031752 | Sushil Dhital | 88

FC7P01NI MSc Project

7.2 Comparison with Related Works

This section evaluates the contributions of this thesis in the context of the related works discussed
in Chapter 2. The comparison highlights the advancements introduced by this research while
addressing the limitations identified in existing studies.

This research builds on and addresses the limitations identified in previous studies, introducing
significant advancements in blood donation management. Unlike earlier works that relied on static
and region-specific datasets, such as those used by (Epifani, et al., 2023), this thesis utilized a
more diverse dataset from multiple regions in Nepal, including Kathmandu, Lalitpur, and
Bhaktapur. This broader scope ensures that the findings are applicable across different geographic
and demographic contexts.

While many existing studies focused on static analysis without real-time functionalities, this
research conceptualized a mobile application prototype with features like geolocation-based donor
tracking, real-time notifications, and dynamic donor-recipient matching. These additions enhance
the responsiveness and practicality of blood donation systems, addressing a critical gap in real-
time applications.

Previous works, such as those by (El-rashidy, et al., 2023), primarily concentrated on

classification tasks and overlooked advanced machine learning techniques. In contrast, this thesis
applied clustering, regression, and predictive analytics, enabling a more comprehensive
understanding of donor behaviour and improving forecasting accuracy. Moreover, while most
studies failed to integrate their findings with mobile platforms, this research bridged that gap by
designing an iOS-based application prototype, demonstrating how analytics can be transformed
into accessible, user-friendly solutions.

Finally, this research emphasized ethical considerations, such as data privacy and equitable access,
which were largely unaddressed in prior studies. By implementing secure data handling protocols
and ensuring transparency, this thesis underscores the importance of ethical compliance in
healthcare technology. These contributions collectively demonstrate the practical and theoretical
advancements introduced by this study.

19031752 | Sushil Dhital | 89

FC7P01NI MSc Project

Chapter 8: Conclusion and Recommendations

This study aimed to address critical challenges in Nepal’s blood donation management system
through a data-driven approach. By leveraging exploratory data analysis, predictive modelling,
and clustering techniques, the research identified demographic trends, geographic disparities, and
donor behaviour patterns, fulfilling the first research question. Urban areas like Kathmandu
exhibited higher donor participation, while rural regions faced significant engagement gaps,
highlighting areas for targeted campaigns. Predictive modelling using techniques like XGBoost
provided valuable insights into donor behaviour, although limitations in accuracy underscored the
need for richer datasets and additional features. Clustering analysis further segmented donors into
actionable groups, offering strategies for targeted mobilization and resource optimization.
Additionally, the design of the BloodConnect app demonstrated how technology could bridge gaps
in donor engagement and emergency responsiveness, laying a foundation for future development.

All research questions were addressed effectively within the scope of this study. The first question,
focused on identifying demographic and geographic disparities, was answered through
comprehensive exploratory data analysis. The second question, exploring the use of predictive and
clustering techniques, was fulfilled by employing machine learning models that provided
actionable insights into donor behaviour and resource allocation. The third question, concerning
the integration of these insights into a mobile application, was addressed conceptually through the
design of the BloodConnect prototype, which incorporated survey feedback to align features with
user needs. While the app remains in the design phase without backend functionality, it
demonstrates the potential for integrating data-driven insights into practical tools.

19031752 | Sushil Dhital | 90

FC7P01NI MSc Project

Chapter 9: References

vaibhav_tyagi, 2024. What are the Principles of Machine Learning? - GeeksforGeeks. [Online]
Available at: [Link]
[Accessed December 2024].

Brodowicz, M., 2024. The Importance of Theoretical Foundation in Scientific Research | Free
Essay Example for Students. [Online]
Available at: [Link]
scientific-research
[Accessed December 2024].

Google Cloud, 2024. What is Supervised Learning? | Google Cloud. [Online]

Available at: [Link]
[Accessed 2024].

Crabtree, M. & Nehme, A., 2024. What is Data Analysis? An Expert Guide With Examples |
DataCamp. [Online]
Available at: [Link]
[Accessed December 2024].

nikhilaggarwal3, 2024. What is Exploratory Data Analysis? - GeeksforGeeks. [Online]

Available at: [Link]
[Accessed December 2024].

Cuemath, 2024. Histogram - Graph, Definition, Properties, Examples. [Online]

Available at: [Link]
[Accessed December 2024].

CFI Team, 2024. Bar Charts - Definition, What It Is, Examples, Types. [Online]
Available at: [Link]
data-visualization-and-technical-analysis/
[Accessed December 2024].

19031752 | Sushil Dhital | 91

FC7P01NI MSc Project

deepanshu_jain, 2024. Understanding Hypothesis Testing - GeeksforGeeks. [Online]

Available at: [Link]
[Accessed January 2025].

Whitfield, B., 2024. Introduction to the Correlation Matrix | Built In. [Online]
Available at: [Link]
[Accessed 2025].

prakharr0y, 2024. T-test - GeeksforGeeks. [Online]

Available at: [Link]
[Accessed 2025].

gajawada, s. k., 2019. ANOVA for Feature Selection in Machine Learning | by sampath kumar
gajawada | Towards Data Science. [Online]
Available at: [Link]
d9305e228476
[Accessed 2025].

ainykhan1410, 2024. What is Predictive Analytics and How does it Work? - GeeksforGeeks.
[Online]
Available at: [Link]
[Accessed December 2024].

avik_dutta, 2024. Random Forest Regression in Python - GeeksforGeeks. [Online]

Available at: [Link]
[Accessed December 2024].

nikki2398, 2023. Gradient Boosting in ML - GeeksforGeeks. [Online]

Available at: [Link]
[Accessed 2025].

Sethi, A., 2024. Support Vector Regression In Machine Learning - Analytics Vidhya. [Online]
Available at: [Link]
for-machine-learning/
[Accessed 2025].

19031752 | Sushil Dhital | 92

FC7P01NI MSc Project

Tyagi, A., 2024. What is the XGBoost algorithm and how does it work?. [Online]
Available at: [Link]
understand-the-math-behind-xgboost/
[Accessed 2025].

Webster, W., 2025. Cluster Analysis/ Definition and Examples - Qualtrics. [Online]
Available at: [Link]
[Accessed 2025].

Javatpoint, 2024. K-Means Clustering Algorithm - Javatpoint. [Online]

Available at: [Link]
[Accessed 2025].

Plutora, 2023. Incremental Model/ Efficient Development Strategies. [Online]

Available at: [Link]
[Accessed 2025].

Apple, 2025. Swift - Apple Developer. [Online]

Available at: [Link]
[Accessed 2025].

Apple, 2025. Xcode - Apple Developer. [Online]

Available at: [Link]
[Accessed 2025].

Epifani, I., Lanzarone, E. & Guglielmi, A., 2023. Predicting Donations and Profiling Donors in a
Blood Donation Environment. Predicting Donations and Profiling Donors in a Blood Donation
Environment.

El-rashidy, N., El-Ghamry, A. & ElSayed, N. E., 2023. Machine Learning for Blood Donors
Classification Model Using Data Mining Techniques. Machine Learning for Blood Donors
Classification Model Using Ensemble Learning.

Selvaraj, P., Sarin, A. & Seraphim, B. I., 2022. Blood Donation Prediction System Using
Machine Learning Techniques. Blood Donation Prediction System Using Machine Learning
Techniques.

19031752 | Sushil Dhital | 93

FC7P01NI MSc Project

Alkahtani, A. S. & Jilani, M., 2019. Predicting Return Donor and Analyzing Blood Donation
Time Series Using Machine Learning Techniques. Predicting Return Donor and Analyzing
Blood Donation Time Series using Data Mining Techniques.

Kumar, M., 2024. iOS App Development/ Understanding Architecture, Swift, and UIKit. [Online]
Available at: [Link]
swift-and-uikit/
[Accessed 2025].

Dahal, R., 2021. Data Analysis/ Five Steps to Superior Data | Grepsr. [Online]
Available at: [Link]
analysis/
[Accessed 2025].

Hamro LifeBank, 2019. report-on-blood-management-system-of-nepal-and-future-possibilities.

[Online]
Available at: [Link]
[Link]
[Accessed 2025].

NRCS, 2025. NRCS - Nepal Red Cross Society. [Online]

Available at: [Link]

19031752 | Sushil Dhital | 94

FC7P01NI MSc Project

Appendix

• Meeting Logs

19031752 | Sushil Dhital | 95

FC7P01NI MSc Project

19031752 | Sushil Dhital | 96

FC7P01NI MSc Project

19031752 | Sushil Dhital | 97

FC7P01NI MSc Project

19031752 | Sushil Dhital | 98

FC7P01NI MSc Project

19031752 | Sushil Dhital | 99

FC7P01NI MSc Project

• Python Scripts for Dashboard Implementation

1. import pandas as pd
2. import streamlit as st
3. from PIL import Image
4. import [Link] as plt
5.
6. # Set page configuration
7. st.set_page_config(layout="wide", page_title="Thesis Results Dashboard")
8.
9. # Add a title at the top of the dashboard
10. [Link](
11. """<h1 style='text-align: center; color: lightcoral;'>
12. Donor Behaviour Analysis and Predictive Modelling for Blood Donation Systems: A
Data-Driven Approach in Nepal's Bagmati Region
13. </h1>""",
14. unsafe_allow_html=True,
15. )
16. [Link](
17. """<p style='text-align: center;'>
18. Explore the key findings and insights derived from this research.
19. </p>""", unsafe_allow_html=True
20. )
21.
22. # Create navigation tabs
23. section = [Link](
24. "Navigate to Section:",
25. ("EDA", "Predictive Modeling", "Model Comparison", "Clustering"),
26. horizontal=True
27. )
28.
29. # Exploratory Data Analysis (EDA) Section

19031752 | Sushil Dhital | 100

FC7P01NI MSc Project

30. if section == "EDA":

31. [Link]("Exploratory Data Analysis (EDA)")
32.
33. # Initialize session state for toggling visualizations
34. if "show_age" not in st.session_state:
35. st.session_state.show_age = False
36. if "show_gender" not in st.session_state:
37. st.session_state.show_gender = False
38. if "show_location" not in st.session_state:
39. st.session_state.show_location = False
40. if "show_blood_group" not in st.session_state:
41. st.session_state.show_blood_group = False
42. if "show_trends" not in st.session_state:
43. st.session_state.show_trends = False
44.
45. # Row 1: Age and Gender Distribution
46. row1_col1, row1_col2 = [Link](2)
47.
48. with row1_col1:
49. if [Link]("Show Age Distribution"):
50. st.session_state.show_age = not st.session_state.show_age
51. if st.session_state.show_age:
52. [Link]("Histogram of Age Distribution")
53. age_image = [Link]("age_distribution.png")
54. [Link](age_image, caption="Age Distribution", use_column_width=True)
55.
56. with row1_col2:
57. if [Link]("Show Gender Distribution"):
58. st.session_state.show_gender = not st.session_state.show_gender
59. if st.session_state.show_gender:
60. [Link]("Bar Plot of Gender Distribution")

19031752 | Sushil Dhital | 101

FC7P01NI MSc Project

61. gender_image = [Link]("gender_distribution.png")

62. [Link](gender_image, caption="Gender Distribution", use_column_width=True)
63.
64. # Row 2: Location and Blood Group Distribution
65. row2_col1, row2_col2 = [Link](2)
66.
67. with row2_col1:
68. if [Link]("Show Location Distribution"):
69. st.session_state.show_location = not st.session_state.show_location
70. if st.session_state.show_location:
71. [Link]("Bar Plot for Location Distribution")
72. location_image = [Link]("location_distribution.png")
73. [Link](location_image, caption="Location Distribution", use_column_width=True)
74.
75. with row2_col2:
76. if [Link]("Show Blood Group Distribution"):
77. st.session_state.show_blood_group = not st.session_state.show_blood_group
78. if st.session_state.show_blood_group:
79. [Link]("Bar Plot for Blood Group Distribution")
80. blood_group_image = [Link]("blood_group_distribution.png")
81. [Link](blood_group_image, caption="Blood Group Distribution",
use_column_width=True)
82.
83. # Row 3: Trends in Blood Groups
84. row3_col1, _ = [Link]([1, 1])
85.
86. with row3_col1:
87. if [Link]("Show Trends in Blood Groups"):
88. st.session_state.show_trends = not st.session_state.show_trends
89. if st.session_state.show_trends:
90. [Link]("Trends in Blood Groups")

19031752 | Sushil Dhital | 102

FC7P01NI MSc Project

91. trends_image = [Link]("blood_group_trends.png")

92. [Link](trends_image, caption="Trends in Blood Groups",
use_column_width=True)
93.
94. # Predictive Modeling Section
95. elif section == "Predictive Modeling":
96. [Link]("Predictive Modeling Results")
97.
98. # List of models with corresponding files
99. models = [
100. ("Linear Regression", "lr_metrics.csv", "lr_coefficients.csv"),
101. ("Random Forest", "rf_metrics.csv", "rf_feature_importance.csv"),
102. ("Gradient Boosting", "gb_metrics.csv", "gb_feature_importance.csv"),
103. ("Support Vector Regression (SVR)", "svr_metrics.csv", None),
104. ("XGBoost", "xgb_metrics.csv", "xgb_feature_importance.csv")
105. ]
106.
107. # Loop through each model and display results
108. for model_name, metrics_file, feature_file in models:
109. with [Link](f"{model_name} Results"):
110. # Load Evaluation Metrics
111. metrics = pd.read_csv(metrics_file)
112. metrics["Value"] = metrics["Value"].round(2)
113.
114. # Load Feature Importance or Coefficients (if applicable)
115. features = None
116. if feature_file:
117. features = pd.read_csv(feature_file)
118. [Link] = [Link]() # Strip any whitespace
119. if "Coefficient" in [Link]: # For Linear Regression
120. features["Coefficient"] = features["Coefficient"].round(2)

19031752 | Sushil Dhital | 103

FC7P01NI MSc Project

121. if "Importance" in [Link]: # For other models

122. features["Importance"] = features["Importance"].round(2)
123.
124. # Display tables side by side
125. col1, col2 = [Link](2)
126.
127. with col1:
128. [Link]("### Evaluation Metrics")
129. [Link](metrics)
130.
131. # Visualize Metrics
132. [Link]("### Metrics Visualization")
133. [Link](figsize=(5, 3))
134. [Link](metrics["Metric"], metrics["Value"], color='lightcoral')
135. [Link](f"{model_name} Metrics")
136. [Link]("Metric")
137. [Link]("Value")
138. [Link](plt)
139.
140. with col2:
141. if features is not None:
142. [Link]("### Feature Importance" if "Importance" in [Link] else
"### Coefficients")
143. [Link](features)
144.
145. # Visualize Feature Importance or Coefficients
146. [Link]("### Feature Visualization" if "Importance" in [Link] else
"### Coefficient Visualization")
147. [Link](figsize=(5, 3))
148. if "Importance" in [Link]:
149. [Link](features["Feature"], features["Importance"], color='skyblue')

19031752 | Sushil Dhital | 104

FC7P01NI MSc Project

150. [Link](f"{model_name} Feature Importance")

151. [Link]("Importance")
152. [Link]("Feature")
153. elif "Coefficient" in [Link]:
154. [Link](features["Feature"], features["Coefficient"], color='mediumseagreen')
155. [Link](f"{model_name} Coefficients")
156. [Link]("Feature")
157. [Link]("Coefficient Value")
158. [Link](plt)
159.
160. # Model Comparison Section
161. elif section == "Model Comparison":
162. [Link]("Model Comparison")
163.
164. # Load comparison metrics
165. comparison_data = {
166. "Model": ["Linear Regression", "Random Forest", "Gradient Boosting", "SVR",
"XGBoost"],
167. "MAE": [4.53, 4.97, 4.50, 4.61, 4.50],
168. "MSE": [33.01, 41.37, 32.84, 42.47, 32.75],
169. "R²": [0.13, -0.09, 0.13, -0.12, 0.14]
170. }
171. comparison_df = [Link](comparison_data)
172.
173. # Highlight XGBoost as the best model
174. [Link]("### Comparison of Predictive Models")
175. [Link](comparison_df)
176. [Link]("**XGBoost performed the best based on the lowest MAE, MSE, and highest
R².**")
177.
178. # Clustering Results Section

19031752 | Sushil Dhital | 105

FC7P01NI MSc Project

179. elif section == "Clustering":

180. [Link]("Clustering Results")
181.
182. # Elbow Curve
183. [Link]("### Elbow Curve for Optimal Clusters")
184. elbow_image = [Link]("elbow_curve.png")
185. [Link](elbow_image, caption="Elbow Curve for Optimal Number of Clusters",
use_column_width=True)
186.
187. # Cluster Summary
188. cluster_summary = pd.read_csv("cluster_summary.csv")
189. [Link]("### Cluster Summary")
190. [Link](cluster_summary)
191.
192. # Cluster Visualization
193. [Link]("### Cluster Visualization")
194. cluster_image = [Link]("kmeans_clusters.png")
195. [Link](cluster_image, caption="K-Means Clustering: Age vs Total Donations",
use_column_width=True)
196.

19031752 | Sushil Dhital | 106

FC7P01NI MSc Project

• Mobile App Wireframes

19031752 | Sushil Dhital | 107

FC7P01NI MSc Project

19031752 | Sushil Dhital | 108

FC7P01NI MSc Project

• Online Meeting requests through Gmail with External Supervisor

• Purchase of Domain for the application

19031752 | Sushil Dhital | 109

FC7P01NI MSc Project

• Managing Thesis Folder

19031752 | Sushil Dhital | 110