Brain Stroke Prediction

Dissertation submitted in fulfilment of the requirements for the Degree of

BACHELORS OF TECHNOLOGY
in
COMPUTER SCIENCE AND ENGINEERING
WITH SPECIALIZATION IN DATA SCIENCE(AI & ML)

By
Garvit Joshi
12106692

Supervisor
Mr. Ved Prakash Chaubey

School of Computer Science and Engineering

Lovely Professional University
Phagwara, Punjab (India)
2021 - 2025
@ Copyright LOVELY PROFESSIONAL UNIVERSITY, Punjab (INDIA)

2024

ALL RIGHTS RESERVED

 DECLARATION STATEMENT

I hereby declare that the research work reported in the dissertation/dissertation

proposal entitled "BRAIN STROKE PREDICTION USING MACHINE
LEARNING” in partial fulfilment of the requirement for the award of Degree for
Bachelor of Technology in Computer Science and Engineering at Lovely Professional
University, Phagwara, Punjab is an authentic work carried out under the supervision
of my research supervisor Mr. Ved Prakash Chaubey Sir. I have not submitted this
work elsewhere for any degree or diploma.

I understand that the work presented herewith is in direct compliance with

Lovely Professional University’s Policy on plagiarism, intellectual property rights,
and the highest standards of moral and ethical conduct. Therefore, to the best of my
knowledge, the content of this dissertation represents an authentic and honest research
effort conducted, in its entirety, by me. I am fully responsible for the contents of my
dissertation work.

Signature of Candidate

Garvit Joshi

12106692
 SUPERVISOR’S CERTIFICATE

This is to certify that the work reported in the B. Tech Dissertation/dissertation

proposal entitled “BRAIN STROKE PREDICTION USING MACHINE
LEARNING”, submitted by Garvit Joshi at Lovely Professional University,
Phagwara, India is a Bonafide record of his / her original work carried out under my
supervision. This work has not been submitted elsewhere to any other degree.

Signature of Supervisor

(Mr. Ved Prakash Chaubey)

Date: 23-04-2024
Counter Signed by:

1) Concerned HOD:
HoD’s Signature:
HoD Name:
Date:

2) Neutral Examiners:

External Examiner

Signature:

Name:

Affiliation:

Date:

Internal Examiner

Signature:

Name:

Date:
 TABLE OF CONTENT

Topic Name Page Number

i) Introduction 6

ii) Problem Statement 7

iii) Working on the Brain Stroke Prediction 8

Model

a) Data Collection 9

b) Data Preprocessing 10

c) Model Development 15

d) Deployment and Future Directions 22

iv) Conclusion 26

v) Bibliography 27
 Checklist for Dissertation-III Supervisor
Name: UID: Domain:
Registration No: Name of student:

Title of Dissertation:

 Front pages are as per the format.

 Topic on the PAC form and title page are same.
 Front page numbers are in roman and for report, it is like 1, 2, 3…….
 TOC, List of Figures, etc. are matching with the actual page numbers
in the report.
 Font, Font Size, Margins, line Spacing, Alignment, etc. are as per the
guidelines.
 Color prints are used for images and implementation snapshots.
 Captions and citations are provided for all the figures, tables etc. and
are numbered and center aligned.
 All the equations used in the report are numbered.
 Citations are provided for all the references.
 Objectives are clearly defined.
 Minimum total number of pages of report is 50.
 Minimum references in report are 30.

Here by, I declare that I had verified the above mentioned points in the
final dissertation report.

Signature of Supervisor with UID

 Abstract

A stroke, often referred to as a cerebrovascular accident (CVA), occurs when a portion

of the brain loses blood flow, which causes the area of the body that those brain cells
control to become dysfunctional. Because of poor blood flow or bleeding into the brain
tissue, this decrease of blood supply may be ischemic or hemorrhagic. Due to the
possibility of fatality or lifelong impairment, a stroke is a medical emergency.
Ischemic strokes may be managed; however, this treatment must begin within a few
hours of the onset of stroke symptoms. If a stroke is suspected, the patient, their
family, or witnesses should call emergency medical assistance right away. A transient
ischemic attack (TIA or mini stroke) is a brief ischemic stroke in which the symptoms
disappear on their own. This situation also demands an immediate evaluation to lower
the risk of a future stroke. If all symptoms go away within 24 hours, that would be a
stroke, not a TIA. Stroke is the second leading cause of mortality globally, claims the
World Health Organization (WHO). accounting for around 11% of all fatalities. Our
ML model uses a dataset for survival prediction to determine a patient's likelihood of
suffering a stroke based on inputs including gender, age, various illnesses, and
smoking status. Our dataset, in contrast to most others, concentrates on characteristics
that would be significant risk factors for a brain stroke.
 Introduction
Brain stroke, also known as a cerebrovascular accident (CVA), is a
medical emergency characterized by the sudden interruption of blood
flow to the brain, leading to a range of neurological impairments.
According to the World Health Organization (WHO), strokes are one of
the leading causes of mortality and long-term disability worldwide. Early
detection and prompt intervention are crucial in preventing the
devastating consequences of strokes and improving patient outcomes.

In recent years, advancements in machine learning and artificial

intelligence have revolutionized healthcare by enabling the development
of predictive models for various medical conditions. Leveraging the
power of machine learning algorithms, researchers and healthcare
professionals have been exploring the potential to predict brain strokes
based on a combination of demographic, clinical, and lifestyle factors.

This project aims to contribute to the ongoing efforts in stroke prevention

and management by developing a machine learning-based predictive
model for brain stroke prediction. By analyzing relevant patient data,
including demographic information, medical history, and lifestyle factors,
the model seeks to identify individuals at risk of experiencing a stroke.
Early identification of high-risk individuals can facilitate timely
interventions, such as lifestyle modifications, medication, or medical
procedures, to mitigate the risk of stroke [Link] this report, we
present the methodology, findings, and implications of our brain stroke
prediction model. We discuss the data collection process, feature
selection, model development, and evaluation metrics used to assess the
performance of the predictive model. Furthermore, we explore the
potential applications of the model in clinical practice and discuss
avenues for future research and improvement.

By harnessing the capabilities of machine learning, this project

endeavours to contribute to the advancement of stroke prevention
strategies and ultimately enhance the quality of care for individuals at risk
of experiencing a brain stroke.
 Problem Statement
1. Significant Burden on Society:

Problem: Brain stroke is a leading cause of death and disability

worldwide. According to the World Health Organization (WHO), stroke
is the second leading cause of death globally, responsible for
approximately 14.4 million deaths each year [WHO, Prevent Brain
Stroke].

Data Impact: This immense death toll translates to a significant loss of

human life and potential. It creates a burden on families and healthcare
systems.

2. Rising Healthcare Costs:

Problem: Stroke survivors often require long-term care and rehabilitation,

leading to substantial healthcare costs. A study published in the Journal of
the American Heart Association found that the total annual cost of stroke
in the United States is estimated to be over $80 billion [Journal of the
American Heart Association, on the cost of stroke care in the US].

Data Impact: These ever-increasing costs strain healthcare budgets and

limit resources available for other medical needs.

3. Devastating Long-Term Effects:

Problem: Stroke can lead to various long-term disabilities, including

paralysis, speech impairments, cognitive decline, and emotional
problems. This significantly impacts a survivor's quality of life and ability
to function independently.

Data Impact: Millions of people live with stroke-related disabilities,

reducing their overall well-being and societal participation.
 Methodology for Brain Stroke Prediction Model
a) Data Collection:

Collect diverse datasets containing demographic, clinical, and lifestyle

factors relevant to stroke prediction. Ensure data quality and integrity by
verifying sources, addressing missing values, and removing duplicates if
any.

b) Data Preprocessing:

Clean and preprocess the collected data by handling missing values,

encoding categorical variables, and scaling numerical features. Perform
exploratory data analysis (EDA) to gain insights into the distribution and
characteristics of the data.

c) Model Development:

Select appropriate machine learning algorithms for stroke prediction,

such as logistic regression, decision trees, random forests, or neural
networks. Design the architecture of the predictive model, including the
number of layers, neurons, and activation functions. Incorporate feature
selection techniques to identify the most relevant predictors for stroke
risk.

d) Model Training:

Split the preprocessed data into training and testing sets to train and
evaluate the performance of the predictive model. Utilize training
algorithms to optimize model parameters and minimize prediction errors.
Validate the model using cross-validation techniques to assess its
robustness and generalizability.

e) Deployment and Future Directions:

Deploy the trained model in a real-world setting, such as a healthcare

facility or online platform, for stroke risk assessment. Monitor the
performance of the deployed model and collect feedback from users for
continuous improvement. Explore future directions for enhancing the
model's predictive accuracy and scalability, such as incorporating
additional features, refining algorithms, or leveraging emerging
technologies like deep learning and reinforcement learning.
 Data Collection
We acquired datasets from Kaggle, encompassing demographic, clinical,
and lifestyle factors, and then utilized Python libraries to preprocess and
analyze the data. Numpy handled numerical computations, pandas
managed data manipulation, and Matplotlib facilitated data visualization.
Numpy supported efficient array operations, pandas enabled tabular data
manipulation, and Matplotlib visualized insights for exploratory analysis.
Data Preprocessing
In the data preprocessing phase of our brain stroke prediction project, we
focus on two essential tasks: Exploratory Data Analysis (EDA) and
handling null values.

Exploratory Data Analysis (EDA):

EDA involves examining the dataset to understand its structure,

distribution, and relationships between variables. Through EDA, we gain
valuable insights into the data that inform subsequent preprocessing steps
and model development. Key aspects of EDA include:

1. Summary Statistics: Computing descriptive statistics such as mean,

median, and standard deviation to understand the central tendency
and dispersion of numerical features.

2. Data Visualization: Creating visualizations such as histograms, box

plots, and scatter plots to visualize the distribution of data, identify
outliers, and explore relationships between variables.

3. Correlation Analysis: Calculating correlation coefficients to

quantify the strength and direction of relationships between
numerical variables.

Handling Null Values:

Null values, or missing values, are common in real-world datasets and

can adversely affect the performance of machine learning models. To
address null values:

1. Identification: We identify the presence of null values in the

dataset using functions like isnull() or info().

2. Imputation: For numerical features, we may impute missing values

using strategies such as mean, median, or interpolation. For
categorical features, we may impute missing values with the mode
or a separate category.

3. Deletion: In cases where null values are too numerous or cannot be

reliably imputed, we may opt to remove rows or columns.
 Correlation in Columns:

1. Age-Storke:

 There is a strong correlation between age and stroke risk.

 Age is a significant factor in the risk of stroke.

2. BMI-Stroke:
 There is a positive correlation between BMI and stroke risk. This
means that as BMI increases, the risk of stroke also increases.

 People with a higher BMI are more likely to experience a stroke

than people with a lower BMI.

3. Glucose_Level-Stroke:

 The graph does not show a clear relationship between glucose level
and stroke. The data points are scattered throughout the graph, and
there is no clear trend.

 The average glucose level in the graph is higher than the average
stroke level. This is because the line labeled "avg_glucose_level" is
higher than the horizontal line labeled "stroke". However, it is
important to note that the graph does not show the range of normal
glucose levels or stroke levels.
4. Heart_Disease-Stroke:

 There is a positive correlation between heart disease and stroke.

This means that as the level of heart disease increases, the density
of stroke also increases.

 Heart disease is a major risk factor for stroke. People with heart
disease are more likely to have a stroke than people without heart
disease.

5. Hypertension-Stroke:

 There is a strong correlation between hypertension and stroke. This

means that as hypertension levels increase, the likelihood of stroke
also increases.

 Hypertension is a major risk factor for stroke. People with

hypertension are more likely to have a stroke than people with
normal blood pressure.
To create some visualizations without writing code:
from pivottablejs import pivot_ui
pivot_ui(df)
Use pivottable, this allows you to create different basic visualizations to
save time and help you analyse clearly.
 Model Development
Before proceeding with model development, it's essential to preprocess
the data and ensure that all features are in a suitable format for analysis.
In our brain stroke prediction project, we convert categorical variables
into a numerical format using a process called encoding. We replace
categorical labels with corresponding numerical values to facilitate model
training.

For example:

a) Gender: 'Male' is encoded as 0, 'Female' as 1, and 'Other' as -1.

Residence Type: 'Rural' is encoded as 0 and 'Urban' as 1.

Work Type: 'Private' as 0, 'Self-employed' as 1, 'Govt_job' as 2, 'children'

as -1, and 'Never_worked' as -2.

We perform this encoding using the replace() function and ensure that the
encoded values are of the appropriate data type, typically unsigned
integers (np.uint8).

Data Splitting for Training and Testing:

After encoding the categorical variables, we proceed to split the dataset

into training and testing sets. This split allows us to train the predictive
model on a portion of the data and evaluate its performance on unseen
data. We use the train_test_split() function from the
sklearn.model_selection module to randomly divide the dataset into
training and testing sets, following an 80-20 ratio (80% for training and
20% for testing). Additionally, we set a random state (random_state=42)
to ensure reproducibility of the split.
Once the data is split, we verify the dimensions of the training and testing
sets using the shape attribute and print the results to confirm the
correctness of the split. This step ensures that the data is properly
partitioned and ready for model training and evaluation.
 1. Logistic Regression:

Description: Logistic regression is a simple yet powerful algorithm

suitable for binary classification tasks like stroke prediction. It provides
interpretable results and works well with linearly separable data.

Accuracy: 0.939335

2. K-Nearest Neighbors (KNN):

Description: KNN is a non-parametric algorithm that classifies data

points based on the majority class among their k nearest neighbors. It is
intuitive, easy to understand, and can capture local patterns in the data.

Accuracy: 0.939335

3. Support Vector Machine (SVM):

Description: SVM is effective in separating classes by finding the optimal

hyperplane that maximizes the margin between them. It works well with
high-dimensional data and is robust to overfitting.

Accuracy: 0.939335
 4. Random Forest:

Description: Random Forest is an ensemble learning algorithm that

combines multiple decision trees to improve predictive performance. It
can handle non-linear relationships and interactions between features and
is robust to overfitting.

Accuracy: 0.940313

5. Naive Bayes:

Description: Naive Bayes is a probabilistic classifier based on Bayes'

theorem. It assumes independence between features, making it
computationally efficient and robust to irrelevant features.

Accuracy: 0.853229
 6. Decision Tree:

Description: Decision trees are interpretable models that partition the

feature space into regions based on feature thresholds. They are robust to
outliers and can capture non-linear relationships in the data.

Accuracy: 0.917808

7. Gradient Boosting Machines (GBM):

Description: GBM builds trees sequentially, correcting errors of the

previous one. It generally provides better accuracy compared to single
decision trees and is robust to outliers.

Accuracy: 0.940313
 8. XGBoost (Extreme Gradient Boosting):

Description: XGBoost is an advanced implementation of gradient

boosting known for its efficiency and scalability. It automatically handles
missing values and is robust to outliers.

Accuracy: 0.933464

9. AdaBoost (Adaptive Boosting):

Description: AdaBoost sequentially corrects mistakes of weak classifiers,

combining multiple weak classifiers to build a strong one. It is effective
in binary classification problems like stroke prediction.

Accuracy: 0.939335
 10. Neural Networks:

Description: Neural networks can capture complex non-linear

relationships in the data and are suitable for large datasets with high-
dimensional features. They can automatically learn feature
representations from the data, making them powerful for stroke prediction
tasks.

Accuracy: 0.939335
Visualization for Summary of Scores of Each Model
 Deployment

What is Streamlit?

Streamlit is an open-source Python library specifically designed to

simplify the creation of data apps. It allows you to rapidly build user
interfaces for your machine learning models and data analysis projects
entirely in Python, without needing expertise in web development
frameworks like HTML, CSS, or Javascript.

Why Streamlit for this project?

Here's why Streamlit is a great choice for deploying your stroke

prediction model:

 Ease of Use: Streamlit focuses on simplicity. You can write

Python code with Streamlit functions to create interactive elements
like buttons, text boxes, and charts. This makes it ideal for data
scientists and machine learning engineers who may not have
extensive web development experience.
 Fast Prototyping: Streamlit allows you to quickly build a
functional prototype of your web app. This lets you get feedback
from users and iterate on the design efficiently.
 Lightweight Deployment: Streamlit apps are lightweight and can
be deployed on various platforms, including the free Streamlit
Community Cloud. This makes it easy to share your app with
others without needing complex server setups.
Features in the Stroke Prediction Page:

Based on the image you described, here are some potential Streamlit
features in your stroke prediction app:

 User Input Widgets: Streamlit offers various widgets like text

boxes, dropdown menus, and radio buttons. These allow users to
input data points like age, gender, and health conditions, which are
then fed into the machine learning models.
 Model Predictions: Streamlit lets you integrate your machine
learning models into the app. When the user submits their data, the
models generate predictions for stroke risk (e.g., "Stroke" or "No
Stroke") and display them in the app.
 Future Directions
1. Integration into Wearable Devices:

Deploying the stroke prediction model into wearable devices, such as

smartwatches, can provide real-time monitoring and alerts for individuals
at risk. By developing an application that continuously monitors vital
signs and other relevant parameters, users can receive alerts when certain
parameters indicate a potential stroke risk. This proactive approach
empowers individuals to take timely preventive measures and seek
medical attention if necessary.

2. Healthcare Software Solutions:

Extending the model into a more precise software solution for healthcare
institutions, insurance companies, and hospitals can enhance stroke risk
assessment and management. By integrating the predictive model into
electronic health record (EHR) systems or dedicated healthcare software
platforms, healthcare providers can leverage the model's insights to
identify high-risk patients, tailor treatment plans, and allocate resources
more efficiently. Insurance companies can also utilize the model to assess
risk profiles accurately and offer personalized insurance plans.

3. Telemedicine and Remote Monitoring:

Incorporating the stroke prediction model into telemedicine platforms and

remote monitoring systems enables remote consultations and proactive
healthcare management. Patients can use mobile applications or web-
based platforms to input relevant health data, and the model can analyze
this data to assess stroke risk remotely. Healthcare providers can then
intervene as necessary, providing guidance, medication adjustments, or
referrals for further evaluation.

4. Continuous Model Improvement:

Continuously refining and updating the predictive model based on new

data, feedback, and advancements in machine learning techniques is
essential for maintaining its accuracy and effectiveness over time.
Collaborating with medical experts, researchers, and data scientists to
gather additional data, incorporate new features, and optimize model
performance ensures that the model remains relevant and reliable in real-
world healthcare settings.

5. Education and Awareness Campaigns:

Conducting education and awareness campaigns to promote stroke

prevention, early detection, and intervention is crucial for reducing the
burden of stroke-related morbidity and mortality. By disseminating
information about stroke risk factors, symptoms, and preventive
measures, individuals can make informed lifestyle choices and seek
medical attention promptly if needed. Integrating the stroke prediction
model into educational materials and public health initiatives can amplify
its impact and empower communities to prioritize stroke prevention and
management.

By pursuing these deployment and future directions, the stroke prediction

model can have a transformative impact on healthcare delivery, enabling
proactive risk assessment, personalized interventions, and improved
outcomes for individuals at risk of stroke.
 Conclusion

The project aimed to rigorously assess the performance of various machine

learning models on a specific dataset, seeking insights into their predictive
capabilities. Across models like Logistic Regression, Support Vector Machine,
K-Nearest Neighbors, AdaBoost, and Neural Network, a consistent performance
emerged, with accuracies averaging around 93-94%. Notably, ensemble methods
such as Random Forest, Gradient Boosting Machines, and AdaBoost showcased
a slight edge over individual models like Decision Tree and Naive Bayes,
highlighting the effectiveness of collective wisdom in enhancing predictive
accuracy. However, caution is warranted due to potential overfitting detected in
certain models, particularly Decision Tree and Naive Bayes, suggesting a need
for further evaluation beyond simple accuracy metrics. Metrics like precision,
recall, and F1-score provide a more nuanced understanding of a model's
generalization performance beyond the training data. These findings underscore
the iterative nature of model evaluation, emphasizing the importance of
continual refinement and the adoption of ensemble techniques to unlock new
frontiers in predictive modeling. In navigating the complexities of model
selection and evaluation, practitioners are guided by a commitment to rigor and
a quest for insights, paving the way for transformative advancements in the field
of machine learning.
 Bibliography
1. World Health Organization. (2020). Stroke. Retrieved from
[Link]
accident/[Link]

2. Kaggle. (n.d.). Stroke Prediction Dataset. Retrieved from

[Link]
stroke-dataset

3. [Link]
4. Learn about Stroke. [(accessed on 25 May 2022)]. Available
online: [Link]
stroke-matters/learn-about-stroke

5. 2. Elloker T., Rhoda A.J. The relationship between social support and
participation in stroke: A systematic review. Afr. J. Disabil. 2018;7:1–9.
doi: 10.4102/ajod.v7i0.357. [PMC free article] [PubMed] [CrossRef] [Google
Scholar]

6. 3. Katan M., Luft A. Seminars in Neurology. Volume 38. Thieme Medical

Publishers; New York, NY, USA: 2018. Global burden of stroke; pp. 208–211.
[PubMed] [Google Scholar]

7. 4. Bustamante A., Penalba A., Orset C., Azurmendi L., Llombart V., Simats A.,
Pecharroman E., Ventura O., Ribó M., Vivien D., et al. Blood biomarkers to
differentiate ischemic and hemorrhagic strokes. Neurology. 2021;96:e1928–
e1939. doi: 10.1212/WNL.0000000000011742. [PubMed] [CrossRef] [Google
Scholar]

8. 5. Xia X., Yue W., Chao B., Li M., Cao L., Wang L., Shen Y., Li X. Prevalence and
risk factors of stroke in the elderly in Northern China: Data from the National
Stroke Screening Survey. J. Neurol. 2019;266:1449–1458.
doi: 10.1007/s00415-019-09281-5. [PMC free article] [PubMed]
[CrossRef] [Google Scholar]