DETECTING PARKINSON’S DISEASE USING

XGBOOST AND RANDOM FOREST CLASSIFIER

A PROJECT REPORT

SUBMITTED BY

SURYA S

In partial fulfilment for the award of the degree

Of

BACHELOR OF TECHNOLOGY

IN

INFORMATION TECHNOLOGY

JEPPIAAR INSTITUTE OF TECHNOLOGY

ANNA UNIVERSITY : CHENNAI 600 025

MAY 2024

1
 ANNA UNIVERSITY : CHENNAI 600 025

BONAFIDE CERTIFICATE

Certified that this project report “DETECTING PARKINSON’S DISEASE

USING XGBOOST AND RANDOM FOREST ALGORITHM” is the
bonafide work of “SURYA S(210620205041)” who carried out the project work
under my supervision

SIGNATURE SIGNATURE
[Link] PRIYA M.E., [Link] B.E.,M.E.,Ph.D.
Assistant Professor Professor
SUPERVISOR HEAD OF THEDEPARTMENT
Information Technology Information Technology
Jeppiaar Institute of Technology Jeppiaar Institute of Technology
Sriperumbudur, Chennai Sriperumbudur, Chennai.

This project report is submitted for viva voce examination held on

INTERNAL EXAMINER EXTERNAL EXAMINER

JEPPIAAR INSTITUTE OF TECHNOLOGY

SRIPERUMBUDUR, CHENNAI – 631 604

2
 ACKNOWLEDGEMENT

We are personally indebted to a number of persons that a complete

acknowledgement would be encyclopedic. First, I love to record my deepest
gratitude to the Almighty Lord and my family

We would like to express our deepest gratitude and respect to Honourable

Col. Dr. JEPPIAAR, M.A., B.L., Ph.D., Chairman for having given the
opportunity to pursue the education in this premier institution

Our sincere thanks and performed sense of gratitude goes to the respected
chairman for all his effort in educating me in a premier institution.

We take this opportunity to thank our Managing Director of this prestigious

institution Dr. N. MARIE WILSON, [Link]. MBA., Ph.D., Jeppiaar Institute
of Technology, for providing facilities in the college premises for carrying out
this project work.

We express our gratitude to our principal Dr. G.R. SURESH and the Head
of the Department, of Computer Science and Engineering, Dr. P. LATHA M.E.,
Ph.D., for their guidance and advise all through the project.

We convey our sincere and in-depth gratitude to our Project Coordinator

and our internal guide Dr. V. SHANMUGA PRIYA M.E., for their valuable
guidance throughout the duration of this project.

We would also like to thank our parents and friends for the support they
extended during this course of the project.

3
 TABLE OF CONTENTS

CHAPTER NO. TITLE [Link]

ABSTRACT 6
LIST OF FIGURES 7
1. INTRODUCTION
1.1 About this project 8
1.2 Scope of the project 9
1.3 Problem Definition
2. SYSTEM ANALYSIS
2.1 EXISTING SYSTEM 10
2.1.1 LITERATURE SURVEY
2.2 PROPOSED SYSTEM 13
2.2.1 ADVANTAGES
3. REQUIREMENT SPECIFICATION
3.1 INTRODUCTION 15
3.2 INPUT DESIGN
3.2.1 OBJECTIVES 16
3.3 REQUIREMENTS
3.3.1 HARDWARE REQUIREMENTS
3.3.2 SOFTWARE REQUIREMENTS 17
3.4 ALGORITHMS AND TOOLS USED
3.4.1 XGBOOST
[Link] INTRODUCTION
[Link] ABOUT THIS ALGORITHM
[Link] INTEGRATION 20
3.4.2 RANDOM FOREST CLASSIFIER
[Link] INTRODUCTION
[Link] ABOUT THIS PROJECT
[Link] INTEGRATION 22
3.4.3 TECHNOLOGIES AND TOOLS
[Link] CV2
[Link] NUMPY
[Link] SKIIMAGE
[Link] IMUTILS
[Link] LABELENCODER
[Link] PANDAS
[Link] UCI IMAGE REPOSITORY
[Link] JUPYTER NOTEBOOK
[Link] MATPLOTLIB
[Link] PACKAGE INSTALLATION
4. SYSTEM DESIGN
4.1 ARCHITECTURE DIAGRAM 36
4.2 FLOW DIAGRAM 37
4.3 SEQUENCE DIAGRAM 38
4.4 XGBOOST ALGORITHM 39
4.5 RANDOM FOREST CLASSIFIER 40

4
5. SYSTEM ARCHITECTURE AND WORKING
OF ALGORITHMS
5.1 SYSTEM ARCHITECTURE 41
5.2 WORK OF XGBOOST ALGORITHM 43
5.3 WORK OF RANDOM FOREST CLASSIFIER 45
6. CODE AND RESULT
6.1 CODE 48
6.2 RESULT ANALYSIS 54
6.3 RESULT SCREENSHOT 56

7. CONCLUSION 58

8. REFERENCES 59

5
 ABSTRACT

Parkinson’s disease affects movement, causes tremors and stiffness, and is a progressive
illness of the central nervous system. It comprises five and affects more than 1 million people
in India each year. There is currently no treatment for this chronic condition. It is a
neurogenerative condition that affects brain cells that make dopamine. This is recognised
using a computer method called XGBoost. A new machine learning method called XGBoost,
is a decision tree-based algorithm Parkinson’s disease is a neurogenerative disorder affecting
dopamine-producing neurons in the brain, and XGBoost is a new Machine learning algorithm
designed for speed and performance. This Python project will create a model using a
XGBClassifier and determine its correctness.

6
 LIST OF FIGURES

FIGURE NO. FIGURE NAME [Link]

4.1 ARCHITECTURE DIAGRAM 36

4.2 FLOW DIAGRAM 37

4.3 SEQUENCE DIAGRAM 38

4.4 XGBOOST ALGORITHM 39

4.5 RANDOM FOREST CLASSIFIER 40

7
 CHAPTER 1

INTRODUCTION

Parkinson’s disease is a progressive disorder of the central nervous system affecting

movement and inducing tremors and stiffness. It has 5 stages to it and affects more than 1
million individuals every year in India. This is chronic and has no cure yet. It is a
neurodegenerative disorder affecting dopamine-producing neurons in the brain.A
neurological condition called Parkinson's disease (PD) affects 60% of persons over 50.
Parkinson's disease (PWP) patients struggle with speech impairment and movement issues,
which makes it difficult for them to travel for appointments for treatment and monitoring.
Early discovery of PD enables treatment, allowing patients to live normal lives. The necessity
to identify PD early, remotely, and correctly is highlighted by the world's aging
[Link], the early non-motor symptoms of Parkinson's disease (PD) may be
slight and brought on by a variety of other illnesses. Thus, it might be difficult to diagnose
PD at an early stage because these symptoms are frequently disregarded. Machine learning
algorithms have been applied for the classification of PD and healthy controls or patients with
comparable clinical presentations to solve these issues and improve the diagnosis and
assessment procedures of [Link] application of machine learning techniques in telemedicine
to identify PD in its early stages is highlighted in this research. To detect this a machine
algorithm called XGboost is used. XGBoost is a new Machine Learning algorithm designed
with speed and performance in mind. XGBoost stands for eXtreme Gradient Boos ng and is
based on decision trees. In this project, we will import the XGBClassifier from the xgboost
library; this is an implementation of the scikit-learn API for XGBoost [Link] this
Python machine learning project, we'll create a model using an XGBClassifier using the
Python packages scikit-learn, numpy, pandas, and xgboost. After loading the data, obtaining
the features and labels, scaling the features, spliting the dataset, and creating an
XGBClassifier, we will determine the model's correctness.

1.1 About this project:

Detecting Parkinson's disease in its early stages is a critical endeavor that can significantly
impact patient outcomes. Leveraging machine learning algorithms such as XGBoost and
Random Forest presents a promising avenue for achieving this goal. In this project, the focus

8
lies on developing a predictive model capable of identifying early signs of Parkinson's
disease based on a variety of input data, ranging from demographic information to motor and
non-motor symptoms associated with the condition.

1.2 Scope of Project:

The scope of this project encompasses the development and implementation of a machine
learning-based system for the early detection of Parkinson's disease. It involves gathering
comprehensive datasets containing demographic information and various motor and non-
motor symptoms associated with the condition. These datasets undergo rigorous
preprocessing, including feature extraction and selection, to distill relevant information
indicative of Parkinson's disease. The project focuses on the selection and training of
XGBoost and Random Forest classifiers, leveraging their capabilities in handling complex
datasets and capturing non-linear relationships. Evaluation metrics are employed to assess the
models' performance, ensuring their effectiveness in distinguishing individuals with
Parkinson's disease from those without. Ethical considerations, such as patient privacy and
potential biases, are carefully addressed throughout the project. Ultimately, the goal is to
deploy the developed system in clinical settings or integrate it into healthcare systems,
contributing to early detection efforts

1.3 Problem Definition:

The problem addressed here is the early detection of Parkinson's disease, a neurodegenerative
disorder characterized by motor and non-motor symptoms that worsen over time. Early
diagnosis is crucial for effective management and treatment, as it allows for interventions that
can slow the progression of the disease and improve the quality of life for patients. The
project aims to leverage machine learning techniques, specifically XGBoost and Random
Forest classifiers, to develop a predictive model capable of identifying individuals at risk of
developing Parkinson's disease based on a range of demographic and symptom-related data.
By accurately identifying early signs of the disease, the project seeks to enable timely
intervention and treatment, potentially leading to better patient outcomes and reducing the
burden on healthcare systems.

9
 CHAPTER 2

SYSTEM ANALYSIS

2.1 EXISTING SYSTEM

2.1.1 LITERATURE SURVEY:

1) Amin ul Haq [Link]. Parkinson's Disease (PD) is the next most common ailment following
Alzheimer's disease. Medical practioners and researchers have struggled to identify
Parkinson's disease correctly and promptly. Traditional approaches based on reduced
characteristics and Statistical Learning have been developed for PD [Link] literature
shows that when DL-based diagnosis algorithms are trained on larger datasets, their
performance accuracy improves.

2)Rohit Lamba [Link]. The authors propose a speech signal-based hybrid Parkinson's disease
diagnosis system. A hybrid system for Parkinson’s disease diagnosis has shown the best
performance with 95.58% accuracy. This result is also better than the recent work found in
the literature.

3. Mahmoud Ragab and [Link] with Parkinson's disease (PD) have changes in
their ability to write, speak, tremble, and flex their muscles. severe degrees are extremely
dangerous as PD patients proceed rigidity, which makes it impossible to move or stand.

4. Parkinson's disease (PD) is a prevalent, irreversible, progressive neurodegenerative

movement disarray. Techniques for machine learning have been utilized more and more in
PD diagnosis. By combining In these applications, multimodal data (imaging and clinical)
might improve PD diagnosis even more.

5) Emuel Daniel Pah [Link]. Parkinson's disease (PD) is a neurodegenerative disorder in which
the substantia nigra region of the brain is affected. PD is diagnosed based on clinical
observation of the symptoms and self-reported functional impairments. Voice assessment has
the advantage that it can be captured and processed without special equipment. People with
Parkinson's disease (PD) often have dysarthria or slurring of the voice. There are differences

10
between the voice parameters of people with PD and healthy participants. Age, gender, and
ethnicity are confounding factors. Computational approaches to support PD diagnosis and
rehabilitation are being actively investigated.

6) Gunjan Pahuja [Link]. Two frameworks, feature-level and modal-level, are presented to
classify the given subjects into PD and healthy. Neuroimaging (T1 weighted MRI scans and
SPECT) and biological (CSF) features are used to form a heterogeneous dataset for deep
learning models to diagnose PD.

7. Majid Aljalal [Link]. Parkinson's disease (PD) is a brain abnormality that affects people all
over the world. Electroencephalography (EEG) is one of the most important PD diagnostic
tools since this disease is linked to the brain. Novel efficient common spa al pa ern-based
approaches for detecting Parkinson's disease are proposed.

8) Ajay Sankar Gullapalli [Link]. Spontaneous Speech (SS) is an efficient tool for the early
detection of diseases like Parkinson's, Alzheimer's, Autism and several other dementia types
in elderly people. Review of speech features and machine learning algorithms is presented to
help in development of non-invasive signal processing techniques.

9) Vyas Tarjni et [Link]'s disease (PD) is a central nervous system neurological

condition. By identifying biomarkers, MRI can help determine how the disease progresses
and potentially pave the way for a future cure. A 2D CNN and a 3D CNN mode were trained
and tested using 318 MRI scans.

10. Quan Changqin and others. Hypokinetic dysarthria affects about 90% of Parkinson's
disease patients. In this research, a novel end-to-end deep learning approach is suggested.
using speech signals for diagnosis. The suggested model pulls out dynamic aspects of me
series utilizing two-dimensional and One-dimensional convolutional neural networks.

11) Tracy John M. et al. The second most common neurodegenerative disease, Parkinson's
disease (PD), may include voice as a deep phenotype, according to this research, chaos on a
global scale. It examined a PD patient database and speech recordings from non-PD
participants that were utilized to isolate paralinguistic elements that were used as inputs for
PD severity prediction using machine learning techniques.

12. H. Gunduz and others. This study proposes two convolutional neural network-based
frameworks for the classification of Parkinson's disease (PD). The initial framework
combines different characteristic sets prior to using them as inputs in a nine-layered News

11
Corp. Deep learning is a capability of the second framework. features obtained through
parallel convolution layers from each feature set, enhancing the discriminative ability of
classifiers.

13. Turker Tuncer et al.'s study suggests a novel technique that uses vowels to automatically
identify Parkinson's disease (PD) by combining minimum average maximum singular value
decomposition (SVD) and the (MAMa) tree. The. 92.46% was the highest classification
accuracy rate [Link] classifiers are used. The suggested approach is prepared to be
evaluated using a sizable database, which can help neurologists in the PD diagnosis.

14) Moro-Velazquez Laureano et al. This review describes the most prevalent characteristics
and machine learning methods for phonatory diagnosis and severity assessment of
Parkinson's [Link] the vocal and articulatory aspects of speech. It is determined that The
phonological and articulatory components are pertinent to the automated PD detection and
severity evaluation.

15) T.J. Wroge and associates. This study investigated the performance of supervised
classification algorithms in correctly diagnosing Parkinson's disease (PD). The types of peak
accuracy of 85% was higher than the typical clinical the movement and diagnosis accuracy of
non-experts (73.8%) specialists in disorders (79.6%). This implies that reliable models that
transform audio information into a diagnostic instrument for Physicians might be able to
provide diagnoses that are more exact and reasonably priced.

16) Imanne [Link] [Link]. This paper proposes a novel intelligent Parkinson detection
system based on deep learning techniques to analyse gait information. It uses 1D
convolutional neural network (1D-Convnet) to build a Deep Neural Network (DNN)
classifier that processes 18 1Dsignals coming from foot sensors measuring the vertical
ground reaction force (VGRF).

17. Jie Mei et al. Clinical signs and observations are the main basis for the diagnosis of
Parkinson's disease (PD), though traditional diagnostic techniques may Expose yourself to
subjectivity. Methods of machine learning been employed to categorize people with
Parkinson's disease and healthy. Patients or controls with comparable clinical presentations in
order to deal with this. An analysis of 209 scholarly articles indicated that machine learning
techniques have a lot of potential. and new biomarkers to be used in clinical judgment,
leading to a more methodical and comprehensive Parkinson's disease (PD) diagnosis.

12
18) Gabriel Solana-Lavalle [Link] This research increases accuracy and reduces the number of
selected vocal features in PD detection using the newest and largest public dataset. Four
classifiers are used to achieve an accuracy of 94.7%, sensitivity of 98.4%, specificity of
92.68%, and precision of 97.22%. The computational complexity is reduced by selecting no
more than 20 features.

19) Zehra Karapinar Senturk [Link]. Parkinson's disease is brought on by damage to the brain
cells that make dopamine, which is essential for movement control, flexibility, and
adaptation. In this research, feature importance and recursive feature elimination approaches
are proposed for a machine learning-based diagnosis of Parkinson's disease. 93.84 percent
accuracy was attained with the fewest voice features.

20. Dr. Eduardo Tolosa, Prof. Over the next 30 years, it is anticipated that the prevalence of
Parkinson's disease, the second most common neurological disease, will quadruple. The
approval of clinical diagnostic standards, the creation and evaluating the prodromal
Parkinson's disease research criteria illness, as well as the identification of genetic subtypes
and genetic. There are very few variations connected to the risk of Parkinson's disease in
latest discoveries. Research on gene cs and imaging are presently a required component of
clinical practice guidelines, and Considerable advancements have been made in the
identification of medical biomarkers. Alzheimer's disease is evolving from a clinical to a
diagnostic entity supported by biomarkers, and novel Therapeutics that alter disease are
currently being developed.

21. This paper proposed that nature inspired WOA algorithm be used followed by RFECV
for feature selection by which 40 most relevant features are selected. The performance of 11
different Classifiers- Logistic Regression, MLP Classifier, KNN, Support Vector Classifier
(SVC-LINEAR, SVC-RBF), Decision Trees Classifier, Random Forest, AdaBoost Classifier,
Gradient Boosting Classifier, Gaussian Process Classifier, Linear Discriminant Analysis
Classifier is analyzed for Holdout (70:30 train-test split) and tenfold cross validation method
for data spliting.

2.2 PROPOSED SYSTEM:

The proposed system involves the development of a machine learning-based framework for
the early detection of Parkinson's disease. This system integrates data collection,

13
preprocessing, model training, and evaluation stages to achieve its objectives. Initially,
comprehensive datasets containing demographic information and various symptoms
associated with Parkinson's disease are collected. These datasets undergo preprocessing steps,
including feature extraction and selection, to distill relevant information indicative of the
condition. Machine learning models, specifically XGBoost and Random Forest classifiers,
are then trained on the preprocessed data to learn patterns and relationships that distinguish
individuals with Parkinson's disease from those without. The trained models are rigorously
evaluated using appropriate metrics to assess their performance and generalization
capabilities. Ethical considerations are carefully addressed throughout the development
process to ensure patient privacy and mitigate potential biases. Ultimately, the proposed
system aims to deploy the developed predictive model in clinical settings or integrate it into
healthcare systems, contributing to early detection efforts and potentially improving patient
outcomes through timely intervention and treatment.

2.2.1 ADVANTAGES

 Early detection of the disease in its early stages

 More Accuracy percent
 High Scalability
 Continuous improvement over time

The system extracts relevant features from the medical images to capture distinctive patterns
and abnormalities associated with Parkinson's disease. Feature extraction techniques such as
texture analysis, shape analysis, and intensity-based features are applied to highlight regions
of interest and quantify disease-related characteristics present in the images. Using the
extracted features and clinical data, machine learning models are developed to classify
individuals as either Parkinson's disease patients or healthy controls. Various classification
algorithms such as support vector machines (SVM), random forests, and deep learning
models are explored and evaluated to identify the most effective approach for disease
detection.

14
The proposed system includes mechanisms for continuous monitoring and improvement
based on feedback from clinicians and ongoing research advancements. Regular updates are
made to the model based on new data and insights, ensuring that it remains accurate and up-
to-date in diagnosing Parkinson's disease. Additionally, collaborations with research
institutions and healthcare organizations enable the system to benefit from interdisciplinary
expertise and contribute to advancements in Parkinson's disease diagnosis and treatment.

CHAPTER 3

REQUIREMENT SPECIFICATION

3.1 INTRODUCTION

The requirements specification is a technical specification of requirements for the software

products. It is the first step in the requirements analysis process it lists the requirements of a
software system including functional, performance and security requirements. The
requirement specification for this project outlines the key functionalities, features, and
constraints of the machine learning-based system aimed at early detection of Parkinson's
disease. It begins with a clear definition of the project's objectives, emphasizing the
importance of accurately identifying individuals at risk of developing Parkinson's disease.
The requirement specification includes detailed criteria for data collection, specifying the
types of demographic information and symptom-related data needed for model training and
evaluation. Overall, the requirement specification serves as a comprehensive guide for the
development and implementation of the machine learning-based system, providing clarity
and direction to project stakeholders.

3.2 INPUT DESIGN:

The input design is the link between the information system and the user. It comprises the
developing specification and procedures for data preparation and those steps are necessary to
put transaction data in to a usable form for processing can be achieved by inspecting the
computer to read data from a written or printed document or it can occur by having people
keying the data directly into the system. The design of input focuses on controlling the
amount of input required, controlling the errors, avoiding delay, avoiding extra steps and

15
keeping the process simple. The input is designed in such a way so that it provides security
and ease of use with retaining the privacy.

3.2.1 OBJECTIVES:

1. Ensuring the the input data covers a wide range of relevant factors including
demographic information, motor and non-motor symptoms, and potentially
biological markers associated with Parkinson's disease.
2. Data is collected accurately to minimize errors and ensure the reliability of the
input features. This may involve using standardized assessment tools and reliable
sources for demographic information.
3. Focus on collecting data that directly contributes to the early detection of
Parkinson's disease. Prioritize features that are known to be associated with the
condition and have been validated in research studies.
4. Implement measures to ensure the quality and reliability of the input data,
including data cleaning, outlier detection, and error correction techniques, to
minimize the impact of noise and inconsistencies on model performance.

3.3 REQUIREMENTS

3.3.1 HARDWARE REQUIREMENTS

 Processor - Multi-core processors like Intel core i5,i7 (or) AMD Ryzen 5,7 is
good to handle many tasks efficiently.
 GPU -Nvidia GPUs (eg.,Geforce GTX,RTX) or AMD GPUs (eg. AMD
Radeon RX) with high memory bandwidth and CUDA cores.
 RAM - Minimum RAM capacity (eg.,8GB) and Sufficient capacity
(16GB,32GB and more).
 Hard Disk - Solid state Drives or NVMe SSDs with minimum capacity upto
256GB and max capacity to 512GB,1TB.
 PSU - Sufficient wattage to power CPU,GPU,RAM and other
components under full

3.3.2 SOFTWARE REQUIREMENTS

16
  XGBOOST-an optimized distributed gradient boosting library designed to be highly
efficient, flexible and portable.
 Python serves as the primary programming language for implementing data
processing, machine learning algorithms, and application development tasks.
 Jupyter Notebook provides an interactive environment for running Python code and
visualizing results, making it well-suited for exploratory data analysis and prototyping
machine learning models.
 OS-provides functions for interacting with the operating system.
 CV2-provides a real-time optimized computer vision library, tools and hardware.
 numpy-for working with arrays
 skimage for image preprocessing.
 for creating interactive visualizations in python.
 sklearn. ensemble for importing random forest classifier.
 imutils-used for basic image processing functions such as translation, rotation, etc.
 Paths-the paths submodule of imutils includes a function to recursively find images
based on a root directory.
 Labelencoder-For encoding target labels with value between 0 and n_classes-1.
 Matplotlib is the Python library for data visualization, offering a variety of plotting
functions and customization options to visualize data distributions, relationships, and
model performance metrics.

3.4 ALGORITHMS AND TOOLS USED

3.4.1 XGBOOST

[Link] INTRODUCTION

XGBoost is a machine learning algorithm known for its efficiency, speed, and
performance in a variety of tasks, including classification, regression, and ranking. It belongs
to the ensemble learning family, specifically gradient boosting, and is widely used in both
research and industry due to its effectiveness in handling structured data.

[Link] ABOUT THIS ALGORITHM

1. Gradient Boosting Ensemble Method:

17
  XGBoost operates on the principle of ensemble learning, where
multiple weak learners (typically decision trees) are combined to create
a strong learner.
 It builds the ensemble sequentially, with each new learner focusing on
the mistakes made by the previous ones.

2. Gradient Boosting Objective:

 The objective of XGBoost is to minimize a loss function by adding
new models that complement the existing ones.
 It uses gradient descent optimization to iteratively improve the model's
predictions.
3. Key Components:
 Weak Learners(Decision Trees): XGBoost uses decision trees as base
learners. These trees are shallow, often referred to as "weak" learners,
to prevent overfitting and enhance generalization.
 Regularization: XGBoost incorporates regularization techniques such
as L1 and L2 regularization to control model complexity and prevent
overfitting.
 Gradient-based Optimization: XGBoost optimizes the loss function
by computing the gradients of the loss with respect to the model's
predictions, allowing for efficient model updates.
 Tree Pruning: XGBoost employs tree pruning techniques to remove
redundant branches and improve model efficiency without sacrificing
predictive accuracy.
4. Parallel and Distributed Computing:
 XGBoost is designed for scalability and efficiency, with support for
parallel and distributed computing on multi-core CPUs and GPUs.
 It utilizes cache-aware optimization and data sharding techniques to
minimize memory usage and accelerate computation.
5. Advanced Features:
 Cross-validation: XGBoost supports built-in cross-validation
capabilities for model selection and hyperparameter tuning.

18
  Early Stopping: It implements early stopping criteria to prevent
overfitting and improve convergence by halting model training when
performance on a validation set no longer improves.

[Link] INTEGRATION

In the Parkinson's disease detection project, XGBoost can be applied to the dataset
after preprocessing and feature engineering steps. By training an XGBoost classifier on
relevant features extracted from patient data, the algorithm can learn to distinguish between
individuals with Parkinson's disease and healthy controls. Additionally, hyperparameter
tuning and cross-validation techniques can be employed to optimize the model's performance
and ensure [Link] employs a gradient descent approach to minimize a loss function and
gradient boosting to enhance the performance of each decision tree individually. Due to its
versatility in terms of the kinds of problems it can be used to, its high level of accuracy,
capacity to handle enormous datasets, and popularity among data scientists. In gradient
boosting decision trees, we combine many weak learners to come up with one strong learner.
The weak learners here are the individual decision trees. All the trees are connected in series
and each tree tries to minimize the error of the previous tree. Due to this sequential
connection, boosting algorithms are usually slow to learn, but also highly accurate. In
statistical learning, models that learn slowly perform better.

 Initializing model with a constant value

 For m = 1 to M , Compute so-called pseudoresiduals

19
  Fit a base learner (or weak learner, e.g. tree) closed under scaling to pseudo-residuals,
i.e. train it using the training set

 Compute multiplier by solving the following one-dimensional optimization

problem

 Update the model

 Output

3.4.2 RANDOM FOREST CLASSIFIER

[Link] INTRODUCTION

Random Forest is a popular machine learning algorithm known for its simplicity, robustness,
and effectiveness in a wide range of classification tasks. It belongs to the ensemble learning
family, specifically bagging, and is widely used in various domains, including healthcare,
finance, and bioinformatics.

[Link] ABOUT THIS ALGORITHM

1. Ensemble Learning with Decision Trees:

 Random Forest operates on the principle of ensemble learning, where
multiple decision trees are trained independently and combined to
make predictions.

20
  Each decision tree is trained on a random subset of the training data,
and the final prediction is made by aggregating the predictions of all
trees.
2. Random Feature Selection:
 In each decision tree, a random subset of features is considered at each
split, ensuring diversity among the trees and reducing the risk of
overfitting.
 This random feature selection process helps in capturing different
aspects of the data and improves the overall model's performance.
3. Bootstrap Aggregating(Bagging):
 Random Forest uses a technique called bagging to create diverse
training sets for each decision tree.
 It samples with replacement from the original training data to generate
multiple bootstrap samples, each of which is used to train a separate
decision tree.
4. Voting or Averaging:
 During prediction, each decision tree in the Random Forest
independently classifies the input data point.
 The final prediction is obtained by either taking a majority vote (for
classification tasks) or averaging (for regression tasks) the predictions
of all trees.
5. Robustness and Generalization:
 Random Forest is less prone to overfitting compared to individual
decision trees, thanks to the ensemble averaging and randomization
techniques.
 It tends to produce robust and stable predictions, even with noisy or
high-dimensional data.
6. Scalability and Efficiency:
 Random Forest can efficiently handle large datasets and high-
dimensional feature spaces due to its parallelizable nature.
 It is computationally efficient and can be trained and evaluated
relatively quickly, making it suitable for real-world applications.

21
[Link] INTEGRATION

In the Parkinson's disease detection project, Random Forest can be applied to the
preprocessed dataset after feature engineering. By training a Random Forest classifier on
relevant features extracted from patient data, the algorithm can learn to distinguish between
individuals with Parkinson's disease and healthy controls. Additionally, hyperparameter
tuning and cross-validation techniques can be employed to optimize the model's performance
and ensure robustness. Mathematics behind Random Forest:

 Regression problem:
In order to determine how your data branches from each node when utilizing the
Random Forest Algorithm to solve regression problems, you must know the mean
squared error (MSE). To determine which branch is the best choice for your forest,
this formula estimates the distance between each node and the expected actual value.
In this case, fi is the value the decision tree returned, and yi is the value of the data
point you are testing at a particular node.

 Classification problem:
We should be aware that the Gini index, or the algorithm used to choose how nodes
on a decision tree branch are ordered, is frequently employed when generating
Random Forests based on categorization data. The Gini of each branch on a node is
calculated using this formula using the class and probability, indicating which branch
is more likely to occur. In this case, c stands for the number of classes, and pi is the
relative frequency of the class you are observing in the dataset.

3.4.3 TECHOLOGIES AND TOOLS

[Link] CV2
22
Python is a popular programming language that has numerous libraries and modules for
different applications. One of these libraries that is commonly used for image and video
processing is OpenCV. The cv2 module is the main module in OpenCV that provides
developers with an easy-to-use interface for working with image and video processing
functions. The OpenCV library (cv2) serves as a valuable resource for image processing
tasks, particularly if the project involves analyzing medical images or incorporating visual
data into the predictive model. OpenCV is a versatile and widely-used library for computer
vision tasks, offering a comprehensive suite of functions and algorithms for image
manipulation, feature extraction, and object detection.

cv2 can be utilized for various purposes such as preprocessing and enhancing medical
images, extracting relevant features from imaging data, and integrating visual information
into the machine learning pipeline. For instance, if the project involves analyzing brain scans
or other medical images to detect signs of Parkinson's disease, cv2 can be used to perform
tasks like image denoising, edge detection, and image segmentation to improve the quality
and interpretability of the input data. Additionally, cv2 provides functionalities for extracting
texture features, shape descriptors, and other visual cues from images, which can be
incorporated as input features for machine learning models.

There are some key functions in cv2:

 ‘[Link]()’ : Loads an image from the file system.

 ‘[Link]()’ :Displays an image in a window.
 ‘[Link]()’:Converts an image from one color space to another.
 ‘[Link]()’:Resizes an image to a specified width and height.
 ‘[Link]()’:Applies Gaussian blurring to reduce noise in the
image.
 ‘[Link]()’:Applies a threshold to convert a grayscale image to a
binary image.
 ‘[Link]()’:Detects edges in an image using the Canny edge detection
algorithm.

23
  ‘[Link]()’:Finds contours (i.e., outlines) of objects in a binary
image.
 ‘[Link]()’:Detects circles in an image using the Hough
transform.
 ‘[Link]()’:Enhances the contrast of an image using histogram
equalization.
 ‘[Link]()’:Applies median filtering to reduce noise while
preserving edges.
 ‘[Link]()’ and ‘[Link]()’:Perform morphological operations to
enhance or suppress image features.
 ‘[Link]()’:Loads a pre-trained Haar cascade classifier for
object detection.
 ‘[Link]()’:Matches a template image within a larger image
to detect instances of the template.
 ‘[Link]()’:Rotates an image by a specified angle.
 ‘[Link]()’:Flips an image horizontally or vertically.
 ‘[Link]()’:Adds various types of noise to an image.
 ‘[Link]()’:Writes text on an image.
 ‘[Link]()’:Draws contours on an image.
 ‘[Link]()’:Saves an image to the file system.

OpenCV (cv2) finds extensive applications across diverse fields due to its robust set of
functionalities in computer vision and image processing. One major application lies in object
detection and recognition, where it's employed to identify and track objects within images
or video streams. This capability is invaluable in fields like surveillance, where it aids in
monitoring activities and detecting anomalies in real-time.

Additionally, OpenCV facilitates image processing and enhancement, offering a suite of

tools for tasks such as resizing, filtering, and color space transformations. These operations
are pivotal in fields like medical imaging, where they assist in improving image quality and
extracting meaningful features for diagnosis and treatment planning. Moreover, OpenCV

24
plays a crucial role in feature detection and description, enabling the identification of key
points and descriptors in images.

This functionality finds applications in tasks like image stitching and image-based
localization, contributing to advancements in fields such as robotics and augmented reality.
Overall, the versatility of OpenCV makes it an indispensable tool across numerous domains,
empowering researchers and developers to tackle complex challenges in computer vision and
beyond.

[Link] NUMPY

NumPy is a fundamental library in Python for numerical computing, providing support for
large, multi-dimensional arrays and matrices, along with a collection of mathematical
functions to operate on these arrays efficiently.

At its core, NumPy's main object is the ‘ndarray’, a multi-dimensional array that enables
efficient storage and manipulation of homogeneous data. NumPy's array operations are
significantly faster than their equivalent operations on Python lists, making it an essential tool
for scientific computing and data analysis tasks.

NumPy offers a wide range of functionalities, including:

1. Array Creation:
‘[Link]()’, ‘[Link]()’, ’[Link]()’, and ‘[Link]()’
to create arrays of various shapes and sizes.
2. Array Operations:
NumPy supports mathematical operations such as addition, subtraction,
multiplication, and division on arrays, both element-wise and matrix-wise.

25
 3. Broadcasting:
NumPy allows operations between arrays of different shapes and sizes through
broadcasting, enabling efficient computation without the need for explicit
looping.
4. Array Indexing and Slicing:
NumPy provides powerful indexing and slicing capabilities to access elements
or subarrays within arrays efficiently.
5. Linear Algebra:
NumPy includes functions for linear algebra operations such as matrix
multiplication (‘[Link]()’), matrix inversion (‘[Link]()’), and
eigenvalue decomposition (‘[Link]()’).
6. Random Number Generation:
NumPy's ‘[Link]’ module offers functions for generating random
numbers and random samples from various probability distributions.
7. Statistical Functions:
NumPy provides a wide range of statistical functions for calculating measures
such as mean, median, standard deviation, variance, correlation, and
percentiles.
8. File Input/Output:
NumPy supports reading and writing data from/to disk in various formats,
including text files, binary files, and NumPy's own ‘.npy’ format
9. Integration with Other Libraries:
NumPy integrates seamlessly with other scientific computing libraries in
Python, such as SciPy, Matplotlib, and pandas, forming the foundation of the
Python scientific computing ecosystem.

NumPy, an essential library in the Python ecosystem, serves as the backbone for numerical
computing and array manipulation tasks across various domains. Its versatile array data
structure, the ndarray, forms the cornerstone of numerous applications. In data science,
NumPy facilitates data handling, manipulation, and preprocessing, playing a pivotal role in
tasks such as cleaning datasets and preparing them for analysis.

26
Moreover, in machine learning, it serves as the primary data representation format for
algorithms and models, enabling efficient numerical computations and operations on large
datasets. Additionally, NumPy finds extensive use in scientific computing, supporting
simulations, mathematical modeling, and statistical analysis across disciplines.

Its array manipulation capabilities are instrumental in fields like image and signal processing,
where it aids in tasks such as image enhancement, filtering, and feature extraction. Beyond
scientific domains, NumPy is employed in financial modeling, bioinformatics, geospatial
analysis, computer vision, and robotics, showcasing its versatility and wide-ranging
applicability. Whether in education, research, or industrial applications, NumPy remains
indispensable for its efficient numerical computations and array operations, empowering
developers and researchers to tackle complex problems with ease and efficiency.

[Link] SKIMAGE

scikit-image (skimage) is a powerful image processing library in Python, offering a

comprehensive collection of algorithms and tools for a wide range of image processing tasks.
Developed on top of NumPy, scikit-image provides a user-friendly interface and efficient
implementations of various image processing techniques.

With scikit-image, users can perform a diverse array of operations such as image filtering,
segmentation, feature extraction, and image restoration. Its extensive collection of functions
enables tasks like edge detection, object detection, texture analysis, and color manipulation,
making it a valuable resource for both beginners and experienced researchers in the field of
image processing.

One of the key strengths of scikit-image lies in its modularity and ease of use. Its well-
documented API and intuitive design allow users to quickly get started with image processing
tasks, while also providing flexibility for advanced customization and experimentation.
Additionally, scikit-image integrates seamlessly with other scientific computing libraries in
Python, such as NumPy, SciPy, and Matplotlib, facilitating a cohesive workflow for image
analysis and visualization.

27
Moreover, scikit-image is actively developed and maintained by a vibrant community of
contributors, ensuring regular updates, bug fixes, and enhancements. This collaborative effort
not only improves the library's performance and reliability but also fosters a supportive
ecosystem for knowledge sharing and innovation in the field of image processing.

In summary, scikit-image is a valuable tool for anyone working with images in Python,
offering a rich set of functionalities, ease of use, and community support. Whether in
academic research, industrial applications, or hobbyist projects, scikit-image empowers users
to explore, analyze, and manipulate images with confidence and efficiency.

[Link] IMUTILS

Imutils is a Python library crafted to streamline a variety of common image processing tasks,
particularly within the OpenCV ecosystem. With its user-friendly interface and robust
functionality, it serves as a versatile tool for developers and researchers alike. One of its
primary utilities lies in image resizing, offering a straightforward method to adjust image
dimensions while maintaining aspect ratio or specifying custom width and height parameters.
This feature is particularly handy for tasks where image size adjustment is necessary, such as
preparing images for input into machine learning models or web applications.

Another notable capability of imutils is image translation, which enables users to shift images
horizontally or vertically. This functionality is invaluable for tasks like image alignment,
where precise positioning of images is required, or when creating composite images or
mosaics. Additionally, imutils simplifies image rotation, allowing users to rotate images by
any desired angle effortlessly. This feature is useful for tasks such as correcting image
orientation or aligning images for comparison or analysis.

Imutils also provides convenient functions for displaying images within OpenCV windows,
enhancing the development process by facilitating easy visualization. With dynamic resizing

28
capabilities, users can view images at optimal sizes, improving workflow efficiency.
Moreover, imutils offers additional convenience functions like resize, rotate, translate, and
grab_contours, which further streamline common image processing operations and enhance
productivity.

Imutils, a Python library designed to simplify image processing tasks, finds applications
across various domains due to its versatility and convenience. Here are some common uses of
imutils:

 Image Resizing
 Image Rotation
 Image Translation
 Displaying Images
 Basic Image Processing Operations
 Alignment and Registration
 Computer Vision and Robotics
 Education and Research

In the Parkinson's disease detection project, imutils, a utility library for image processing
tasks in Python, can play a significant role in streamlining various aspects of image
manipulation and analysis. One crucial aspect where imutils can be beneficial is in
preprocessing the medical images used for diagnosis. With imutils, tasks like resizing,
rotating, and adjusting the orientation of the images can be performed efficiently. For
instance, medical imaging datasets often contain images of varying sizes and orientations,
and standardizing them is essential for accurate analysis. Imutils simplifies this process by
providing functions to resize and rotate images, ensuring consistency across the dataset.

Furthermore, imutils can aid in extracting relevant features from the medical images, which
are crucial for detecting signs of Parkinson's disease. Features such as edges, textures, and
shapes can be extracted using imutils functions, facilitating subsequent analysis and

29
classification. For example, imutils provides convenient methods for edge detection and
contour extraction, which can help identify distinctive patterns or abnormalities indicative of
Parkinson's disease in brain scans or other medical images.

Python, enhancing the efficiency and Moreover, imutils can streamline the visualization of
images during the development and evaluation of the disease detection models. With its
functions for displaying images in OpenCV windows with dynamic resizing capabilities,
developers can easily inspect the preprocessing steps and the effects of various image
enhancements. This aids in fine-tuning the preprocessing pipeline and ensuring that the input
data fed into the machine learning models is appropriately processed and standardized.

imutils can be utilized for tasks such as image augmentation, which is essential for increasing
the diversity of the dataset and improving the robustness of the trained models. By applying
transformations such as rotation, translation, and flipping to the images, imutils enables the
generation of augmented versions of the data, augmenting the training dataset and reducing
the risk of overfitting.

imutils serves as a valuable tool in the Parkinson's disease detection project, facilitating
image preprocessing, feature extraction, visualization, and data augmentation. Its intuitive
interface and comprehensive functionality make it an indispensable asset for developers and
researchers working with medical images ineffectiveness of image processing tasks.

[Link] LABELENCODER

The LabelEncoder is a utility class in the scikit-learn library, primarily used for encoding
categorical labels into numerical representations. In many machine learning algorithms, it's
essential to convert categorical data into numerical format for model training. LabelEncoder
achieves this by assigning a unique integer to each category present in the dataset.

30
Here's how LabelEncoder works:

1. Fit:
 The LabelEncoder first learns the unique categories present in the data
during the fitting stage.
 It assigns a unique integer to each category encountered, starting from
0 up to the number of distinct categories minus one.

2. Transform:
 Once the LabelEncoder has learned the categories, it transforms the
categorical labels into their corresponding numerical representations
during the transformation stage.
 Each category is replaced with its assigned integer label.

LabelEncoder is particularly useful for encoding target labels in classification tasks, where
the target variable is categorical. It ensures that the model can interpret and process the labels
effectively during training. However, it's important to note that LabelEncoder should only be
used for encoding target labels and not for encoding features, as it may introduce unintended
ordinal relationships between categories.

Here's a simple example demonstrating how to use the LabelEncoder from the scikit-learn
library:

INPUT:

from [Link] import LabelEncoder

# Sample categorical labels

labels = ['cat', 'dog', 'cat', 'bird', 'dog']

# Initialize LabelEncoder

label_encoder = LabelEncoder()

31
# Fit the LabelEncoder to learn the unique categories

label_encoder.fit(labels)

# Transform the categorical labels into numerical representations

encoded_labels = label_encoder.transform(labels)

# Display the original labels and their encoded representations

print("Original Labels:", labels)

print("Encoded Labels:", encoded_labels)

# Optionally, inverse transform the encoded labels back to original labels

decoded_labels = label_encoder.inverse_transform(encoded_labels)

print("Decoded Labels:", decoded_labels)

OUTPUT:

Original Labels: ['cat', 'dog', 'cat', 'bird', 'dog']

Encoded Labels: [0 1 0 2 1]

Decoded Labels: ['cat' 'dog' 'cat' 'bird' 'dog']

In this example,we have a list of categorical labels ‘[ ‘cat’ , ‘dog’ , ‘cat’ , ‘bird’ , ‘dog’]’. We
initialize a LabelEncoder object. We fit the LabelEncoder to the list of labels to learn the
unique categories. We transform the original labels into numerical representations using the
‘transform’ method. We print the original labels and their encoded representations.
Optionally, we can use the ‘inverse_transform’ method to decode the encoded labels back to
their original form.

[Link] PANDAS

The open-source Python data manipulation and analysis toolkit Pandas offers strong and
adaptable data structures including Series and DataFrame. The ability to read and write data
from a range of file types is also built-in. It can work with other Python packages, making it a

32
potent tool for modeling and data analysis. Pandas is a Python module used to expedite data
cleaning, pre-processing, and analysis. Numpy, Python's mathematical library, is the
foundation upon which Pandas is [Link] pandas library enables you to work with
tabular data with columns of various data types, including those from Excel spreadsheets,
CSV files downloaded from the internet, SQL database tables, Time series data, either at
fixed-frequency or not, and other structured datasets, including those derived from web data,
like JSON files

[Link] UCI MACHINE REPOSITORY

A well-known online repository called the UCI Machine Learning Repository offers a variety
of datasets and other resources for testing and comparing machine learning methods. It is free
to use and is managed by the University of California, [Link] provides the machine learning
community with 622 datasets that are currently [Link] machine learning community
uses it as a collection of databases, domain theories, and data generators for the empirical
investigation of machine learning algorithms.

Advantages of UCI Machine Repository:

 Wide Variety of Datasets

 Standardized Format
 Benchmarking
 Educational Resource
 Community Contribution

Disadvantages of UCI Machine Repository:

 Limited Size and Scope

 Data Quality
 Lack of Updates
 Limited Support

33
[Link] JUPYTER NOTEBOOK

Jupyter Notebook is a powerful tool that provides an interactive computing environment for
data analysis, machine learning, and scientific computing tasks. It allows users to create and
share documents that contain live code, equations, visualizations, and narrative text. One of
the key advantages of Jupyter Notebook is its ability to combine code execution, explanatory
text, and visualizations in a single document, making it ideal for exploratory data analysis and
prototyping machine learning models.

With Jupyter Notebook, users can write and execute Python code in a step-by-step manner,
allowing for iterative development and quick experimentation. The ability to run code cells
individually and see immediate results fosters an interactive and exploratory workflow,
enabling users to test hypotheses, visualize data distributions, and debug code in real-time.
Moreover, Jupyter Notebook supports various programming languages, including Python, R,
and Julia, making it versatile for different data analysis [Link] benefit of Jupyter
Notebook is its rich support for data visualization. Users can easily create plots, charts, and
graphs using libraries such as Matplotlib, seaborn, and Plotly, and embed them directly within
the notebook alongside code and text. This seamless integration of code and visualizations
enhances the interpretability of results and facilitates communication of findings to
[Link], Jupyter Notebook promotes collaboration and reproducibility by
allowing users to share their analyses as interactive documents or export them to different
formats such as HTML, PDF, or slides. This enables researchers, data scientists, and analysts
to collaborate on projects, review code, and reproduce results in a transparent and accessible
manner.

[Link] MATPLOTLIB

Matplotlib is a widely-used Python library for creating static, interactive, and animated
visualizations in data analysis, scientific computing, and machine learning projects. It
provides a flexible and comprehensive set of tools for creating a wide range of plots, charts,
and graphs, allowing users to visualize data distributions, relationships, trends, and patterns
effectively. One of the key strengths of Matplotlib is its versatility and customization options.

34
Users can create a variety of plot types, including line plots, scatter plots, bar plots,
histograms, heatmaps, and more, with full control over plot aesthetics such as colors, line
styles, markers, and labels. This flexibility enables users to tailor visualizations to their
specific needs and preferences, ensuring clarity and effectiveness in conveying insights from
the data.

Matplotlib's object-oriented interface allows for fine-grained control over every aspect of a
plot, making it suitable for creating complex and publication-quality figures. Users can create
multi-panel figures, annotate plots with text and arrows, add legends and colorbars, and
customize axis scales and ticks, among other advanced features. Additionally, Matplotlib
seamlessly integrates with other Python libraries such as NumPy, pandas, and scikit-learn,
enabling easy data manipulation and visualization workflows.

Another advantage of Matplotlib is its extensive documentation and large user community.
The library is well-documented with comprehensive tutorials, guides, and examples, making
it accessible to users of all levels, from beginners to experienced developers. Moreover, the
active user community provides support, shares best practices, and contributes to the
development of new features and enhancements, ensuring that Matplotlib remains a robust
and reliable tool for data visualization.

In addition to creating static plots, Matplotlib also supports interactive plotting capabilities
through integration with libraries such as ‘IPython widgets’ and ‘mpld3’. These features
enable users to create interactive visualizations that allow for exploration and manipulation of
data directly within the plot, enhancing the interactive data analysis experience.

[Link] PACKAGE INSTALLATION

To install packages within a Jupyter Notebook, you can use the !pip command followed by
the package name. Here's how you can install packages in a Jupyter Notebook cell:

35
# Install a package

!pip install package_name

Replace package_name with the name of the package you want to install. You can also
specify a specific version of the package by adding ==version_number after the package
[Link] example:

# Install scikit-learn

!pip install scikit-learn

# Install a specific version of pandas

!pip install pandas==1.3.0

CHAPTER 4

SYSTEM DESIGN

4.1 ARCHITECTURE DIAGRAM

36
 FIG.1 ARCHITECTURAL DIAGRAM

4.2 FLOW DIAGRAM

37
 FIG.2 FLOW DIAGRAM

4.3 SEQUENCE DIAGRAM

38
 FIG.3 SEQUENCE DIAGRAM

4.4 XGBOOST ALGORITHM

39
 FIG.4 XGBOOST ALGORITHM

4.5 RANDOM FOREST CLASSIFIER

40
FIG.5 RANDOM FOREST CLASSIFIER

CHAPTER 5

41
 SYSTEM ARCHITECTURE AND WORKING OF
ALGORITHMS

5.1 SYSTEM ARCHITECTURE

The system architecture for the Parkinson's disease detection project comprises several
interconnected components designed to facilitate the efficient processing and analysis of
medical imaging data for accurate disease diagnosis. Here's an overview of the system
architecture:

1. Data Collection and Storage:

The system begins with the collection of medical imaging data from patients
diagnosed with Parkinson's disease and healthy individuals. This data, along
with demographic and clinical information, is stored in a centralized data
repository. The repository should be scalable and secure, capable of handling
large volumes of imaging data while ensuring compliance with privacy
regulations

Image [Link] Images in [Link] Images in Total

Training Set Test Set
Spiral 72 30 102
Wave 72 30 102

2. Preprocessing Module:
Before analysis, the collected imaging data undergoes preprocessing to ensure
consistency and quality. This module includes tasks such as normalization,
denoising, registration, and segmentation to standardize the data and enhance
its suitability for subsequent analysis. Preprocessing algorithms and techniques
may vary depending on the imaging modality and specific requirements of the
dataset.
3. Feature Extraction and Selection:

42
 Once preprocessed, the imaging data is fed into a feature extraction module
where relevant features are extracted to capture disease-related patterns and
characteristics. Feature extraction techniques such as texture analysis, shape
analysis, and intensity-based features are applied to highlight discriminative
information present in the images. Feature selection methods may also be
employed to reduce dimensionality and remove irrelevant or redundant
features, enhancing model efficiency and performance.
4. Machine Learning Model Development:
Extracted features are used to train machine learning models for Parkinson's
disease classification. Various classification algorithms such as support vector
machines (SVM), random forests, and deep learning architectures are explored
and evaluated to identify the most effective approach for disease detection.
The model development process involves training on labeled data, validation
using cross-validation techniques, and optimization of hyperparameters to
maximize performance.
5. Model Evaluation and Validation:
Developed models are rigorously evaluated using independent test datasets to
assess their accuracy, sensitivity, specificity, and robustness. Performance
metrics such as area under the receiver operating characteristic (ROC) curve
are calculated to quantify the models' diagnostic capabilities. Validation
procedures ensure that the models generalize well to unseen data and perform
reliably across different patient cohorts and imaging modalities.
6. Clinical Integration and Deployment:
Once validated, the trained models are integrated into clinical workflows to
assist healthcare professionals in diagnosing Parkinson's disease. A user-
friendly interface allows clinicians to input patient data and medical images,
and the system delivers real-time diagnostic predictions and confidence scores.
Clinicians can interpret the model's predictions in conjunction with other
clinical findings to make informed decisions regarding patient care and
management.

7. Continuous Monitoring and Improvement:

43
 The system includes mechanisms for continuous monitoring and improvement
based on feedback from clinicians and ongoing research advancements.
Regular updates are made to the models based on new data and insights,
ensuring their accuracy and relevance in diagnosing Parkinson's disease.
Collaboration with research institutions and healthcare organizations enables
the system to benefit from interdisciplinary expertise and contribute to
advancements in Parkinson's disease diagnosis and treatment.

5.2 WORKING OF XGBOOST ALGORITHM

The proposed methodology for integrating the XGBoost algorithm into the Parkinson's
disease detection project involves a systematic approach aimed at optimizing model
performance and accuracy. Here's a detailed outline of the methodology:

- The dataset used in here is the Parkinson's Drawing Dataset from Kaggle . The dataset
includes Spiral and Waves drawings created by healthy and Parkinson's disease patients..

- Ensure the dataset is properly cleaned and preprocessed, handling missing values, outliers,
and inconsistencies.

- Standardize the format and resolution of the imaging data to ensure uniformity across the
dataset.

- Extract relevant features from the medical images using techniques such as texture
analysis, shape analysis, and intensity-based features.

- Explore advanced feature extraction methods, including deep learning-based approaches,

to capture high-level features indicative of Parkinson's disease.

- Split the dataset into training, validation, and test sets to facilitate model development and
evaluation.

- Use stratified sampling to ensure a balanced distribution of Parkinson's disease cases and
healthy controls across the datasets.

44
 - Implement the XGBoost algorithm for Parkinson's disease classification, considering its
ability to handle complex datasets and nonlinear relationships.

- Experiment with different hyperparameters such as learning rate, maximum depth, and
regularization parameters to optimize model performance.

- Utilize techniques like early stopping to prevent overfitting and improve convergence
during training.

- Train the XGBoost model on the training dataset using the selected hyperparameters and
feature set.

- Monitor the training process to track performance metrics such as loss function value,
accuracy, and validation error.

- Apply techniques like cross-validation to assess model generalization and robustness to

unseen data.

- Evaluate the trained XGBoost model using the validation dataset to assess its performance
metrics such as accuracy, sensitivity, specificity, and area under the ROC curve.

- Fine-tune the model hyperparameters based on the validation results to further optimize
performance.

- Compare the performance of the XGBoost model with other machine learning algorithms
to identify the most effective approach for Parkinson's disease detection.

- Interpret the XGBoost model to understand the importance of different features in

predicting Parkinson's disease.

- Visualize feature importance scores and partial dependence plots to gain insights into the
model's decision-making process.

- Integrate the trained XGBoost model into a user-friendly interface or clinical decision
support system for deployment in clinical settings.

- Develop a web-based or mobile application that allows healthcare professionals to input

patient data and medical images and receive real-time diagnostic predictions.

- Ensure compliance with regulatory requirements and patient privacy guidelines when
deploying the model in clinical practice.

45
 - Monitor the performance of the deployed XGBoost model in real-world clinical settings
and collect feedback from healthcare professionals.

- Incorporate new data and insights into model retraining and updates to improve diagnostic
accuracy and reliability.

- Collaborate with healthcare institutions and research organizations to validate the model
on diverse patient populations and imaging [Link] following this proposed
methodology, the XGBoost algorithm can be effectively integrated into the Parkinson's
disease detection project, providing accurate and reliable diagnostic predictions for improved
patient care and management.

5.3 WORKING OF RANDOM FOREST CLASSIFIER

The proposed methodology for incorporating the Random Forest classifier algorithm into my
project is designed to systematically address the complexities of medical imaging data
analysis while optimizing diagnostic accuracy.

46
Rigorous preprocessing ensures dataset cleanliness and standardization, essential for accurate
model training. Feature engineering techniques are then employed to extract relevant
information from the imaging data, capturing distinctive patterns indicative of Parkinson's
disease.

The dataset is split into training, validation, and test sets to facilitate model development and
evaluation, with stratified sampling ensuring balanced representation across classes. The
Random Forest classifier is then implemented, leveraging its ensemble of decision trees to
handle complex data structures and nonlinear relationships effectively.

Hyperparameter tuning and cross-validation techniques fine-tune the model for optimal
performance, while model interpretation aids in understanding the features driving diagnostic
predictions.

Once validated, the trained Random Forest model is deployed into clinical workflows,
providing real-time diagnostic predictions to assist healthcare professionals in making
informed decisions.

Continuous monitoring and improvement strategies ensure the model remains accurate and
relevant, contributing to advancements in Parkinson's disease diagnosis and treatment.
Through this comprehensive methodology, the Random Forest classifier becomes a valuable
tool in enhancing diagnostic capabilities and improving patients outcome.

47
Evaluation The proposed system's effectiveness is assessed using the following performance
measures: classification accuracy, sensitivity, and specificity, denoted by the terms TP as true
positive, FP as false positive, TN as true negative, and FN as false negative.

The percentage of correctly identified PD subjects is determined by the sensitivity or true

positive rate.

The percentage of correctly identified healthy or non-PD subjects is determined by the

specificity.

48
 CHAPTER 6

CODE AND RESULT

6.1 CODE

!pip install opencv-python

!pip install xgboost

!pip install scikit-learn

import os

import cv2

import numpy as np

from skimage import feature

import random

import [Link] as plt

from xgboost import XGBClassifier

from [Link] import RandomForestClassifier

[Link]()

get_ipython().run_line_magic('matplotlib', 'inline')

def quantify_image(image):

features = [Link](image, orientations=9,

pixels_per_cell=(10, 10), cells_per_block=(2, 2),

transform_sqrt=True, block_norm="L1")

return features

def load_split(path):

49
# grab the list of images in the input directory, then initialize

# the list of data (i.e., images) and class labels

imagePaths = list(paths.list_images(path))

data = []

labels = []

# loop over the image paths

for imagePath in imagePaths:

# extract the class label from the filename

label = [Link]([Link])[-2]

# load the input image, convert it to grayscale, and resize

# it to 200x200 pixels, ignoring aspect ratio

image = [Link](imagePath)

image = [Link](image, cv2.COLOR_BGR2GRAY)

image = [Link](image, (200, 200))

# threshold the image such that the drawing appears as white

# on a black background

image = [Link](image, 0, 255,

cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)[1]

# quantify the image

features = quantify_image(image)

# update the data and labels lists, respectively

[Link](features)

[Link](label)

return ([Link](data), [Link](labels))

50
!pip install imutils

from imutils import paths

from [Link] import LabelEncoder

from [Link] import RandomForestClassifier

from xgboost import XGBClassifier

from [Link] import confusion_matrix

def train_models(dataset):

# initialize the models

models = {

"Rf": {

"classifier": RandomForestClassifier(random_state=1),

"accuracy": 0,

"sensitivity": 0,

"specificity": 0,

},

"Xgb": {

"classifier": XGBClassifier(),

"accuracy": 0,

"sensitivity": 0,

"specificity": 0,

# define the path to the testing and training directories

path = "C:\\Users\\Admin\\Downloads\\Dataset\\drawings\\" + dataset

51
 trainingPath = [Link]([path, "training"])

testingPath = [Link]([path, "testing"])

# load the data

(trainX, trainY) = load_split(trainingPath)

(testX, testY) = load_split(testingPath)

# encode the labels

le = LabelEncoder()

trainY = le.fit_transform(trainY)

testY = [Link](testY)

# train each model and calculate its metrics

for model in models:

models[model]["classifier"].fit(trainX, trainY)

predictions = models[model]["classifier"].predict(testX)

cm = confusion_matrix(testY, predictions).ravel()

tn, fp, fn, tp = cm

models[model]["accuracy"] = (tp + tn) / float([Link]())

models[model]["sensitivity"] = tp / float(tp + fn)

models[model]["specificity"] = tn / float(tn + fp)

return models

spiralModels = train_models('spiral')

waveModels = train_models('wave')

52
print("Random Forrest vs XGBoost Classifier\n\n")

for metric in ("accuracy", "sensitivity", "specificity"):

print(f"{[Link]()}: ")

print("Random Forrest={:.2f}%, XGBoost={:.2f}% \n".format(

spiralModels['Rf'][metric]*100, spiralModels['Xgb'][metric]*100))

print("Random Forrest vs XGBoost Classifier\n\n")

for metric in ("accuracy", "sensitivity", "specificity"):

print(f"{[Link]()}: ")

print("Random Forrest={:.2f}%, XGBoost={:.2f}% \n".format(

waveModels['Rf'][metric]*100, waveModels['Xgb'][metric]*100))

def test_prediction(model, testingPath):

# get the list of images

testingPaths = list(paths.list_images(testingPath))

output_images = []

# pick 15 images at random

for _ in range(10):

image = [Link]([Link](testingPaths))

output = [Link]()

output = [Link](output, (128, 128))

# pre-process the image

image = [Link](image, cv2.COLOR_BGR2GRAY)

53
 image = [Link](image, (200, 200))

image = [Link](image, 0, 255,

cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)[1]

# quantify the image and make predictions based on the extracted features

features = quantify_image(image)

preds = [Link]([features])

label = "Parkinsons" if preds[0] else "Healthy"

# draw the colored class label on the output image and add it to

# the set of output images

color = (58,228,132) if label == "Healthy" else (255, 0, 0)

[Link](output, label, (3, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.5,

color, 2)

output_images.append(output)

[Link](figsize=(20, 20))

for i in range(len(output_images)):

[Link](5, 5, i+1)

[Link](output_images[i])

[Link]("off")

[Link]()

testingPath = [Link](['C:\\Users\\Admin\\Downloads\\Dataset\\drawings\\spiral',
'testing'])

test_prediction(spiralModels['Rf']['classifier'], testingPath)

testingPath = [Link](["C:\\Users\\Admin\\Downloads\\Dataset\\drawings\\wave",
"testing"])

test_prediction(waveModels['Rf']['classifier'], testingPath)

54
6.2 RESULT ANALYSIS
In this project, two machine learning classifiers, Random Forest and XGBoost, were trained
and evaluated on two different datasets: 'spiral' and 'wave'. The classifiers were assessed
based on three metrics: accuracy, sensitivity, and specificity, which provide valuable insights
into the performance of the models. Here's a detailed analysis of the results:

Spiral Dataset:
Random Forest vs. XGBoost:
Accuracy: Random Forest outperformed XGBoost with an accuracy of 90.91% compared to
72.73%. This indicates that Random Forest made accurate predictions for a higher percentage
of the test data points.

Sensitivity: Random Forest demonstrated higher sensitivity (86.67%) than XGBoost

(66.67%). This means that Random Forest identified a larger proportion of true positive cases
(Parkinson's patients) compared to XGBoost
.
Specificity: Random Forest exhibited higher specificity (94.44%) compared to XGBoost
(77.78%). This indicates that Random Forest identified a larger proportion of true negative
cases (healthy individuals) compared to XGBoost.

Overall Analysis (Spiral Dataset): Random Forest: Random Forest showed superior
performance on the Spiral dataset, accurately identifying both Parkinson's patients and
healthy individuals. It exhibited high accuracy, sensitivity, and specificity, making it a robust
choice for this dataset.

XGBoost: While XGBoost performed reasonably well, especially in terms of specificity, it

lagged behind Random Forest in accuracy and sensitivity. This suggests that XGBoost might
need further tuning or feature engineering to match the performance of Random Forest on
this dataset.

Wave Dataset:
Random Forest vs. XGBoost:

Accuracy: Random Forest achieved an accuracy of 76.67%, while XGBoost had an accuracy
of 73.33%. Random Forest performed slightly be er in terms of overall accuracy.

55
Sensitivity: XGBoost outperformed Random Forest in sensitivity, with a score of 80.00%
compared to 73.33%. This indicates that XGBoost identified a higher percentage of true
positive cases in the Wave dataset.

Specificity: Random Forest showed higher specificity (80.00%) than XGBoost (66.67%).
This suggests that Random Forest identified a larger proportion of true negative cases
compared to XGBoost.

Overall Analysis (Wave Dataset): Random Forest: Random Forest demonstrated balanced
performance on the Wave dataset, with competive accuracy, sensitivity, and specificity
scores. It successfully identified both Parkinson's patients and healthy individuals.

XGBoost: XGBoost showed strong sensitivity, indicating its ability to detect Parkinson's
cases effectively. However, it lagged behind Random Forest in terms of overall accuracy and
specificity. Further opmization and feature engineering might enhance its performance on this
dataset.

Conclusion: Random Forest vs. XGBoost: Random Forest consistently outperformed

XGBoost in accuracy and specificity on both datasets. It proved to be a robust and reliable
choice for this classification task, achieving high accuracy while effectively identifying both
positive and negative cases.

Areas for Improvement: While Random Forest performed well, there is room for
improvement in the sensitivity of the models, especially for XGBoost. Fine-tuning
hyperparameters, experimenting with different feature extraction techniques, or exploring
additional features could potentially enhance the models' ability to identify Parkinson's
disease cases more accurately.

Further Research: Exploring advanced feature engineering techniques, conducting a more in

depth analysis of misclassified instances, and experimenting with ensemble methods could be
valuable directions for future research to further improve the models' performance.

In summary, Random Forest emerged as the be er-performing classifier for this specific task,
demonstrating its effectiveness in accurately diagnosing Parkinson's disease based on the
given features. However, continuous refinement and exploration of techniques are essential to
enhance the models' sensitivity and overall performance further.

56
6.3 RESULT SCREENSHOT

57
58
 CHAPTER 7

CONCLUSION

Parkinson's disease, which affects the central nervous system (CNS) of the brain, is fatal if
not detected early. There are various stages of Parkinson’s. If it is found in early stages it is
possible to cure or extend life. Due to late finding, treatment is not provided, and lives are
lost. It is crucial to make an early diagnosis as a result. For early disease diagnosis, we
employed machine learning techniques like XGBoost and Random Forest classifiers.
According to our analysis of our data on Parkinson's disease, the algorithm known as
XGBoost is the most accurate in identifying the condition's onset. XGBoost finds the status
of whether a person has Parkinson's or not by evaluating the dataset of numeric values such
as Mdvp are given and it gives the accuracy of how correctly it predicts the outcome. While
the Random Forest classifier uses image dataset, where the patient is required to draw a spiral
and wave drawing. The Random Forest classifier then differentiates the people who have
Parkinson's and those who don’t have Parkinson’s. The output is labelled as either
Parkinson’s or Healthy. With the help of Machine learning techniques, we can train the model
to be more accurate which in turn gives be er results. Early treatment will be possible as a
result, perhaps saving lives.

59
 CHAPTER 8

REFERENCES
[1] .Lamba, Rohit, et al. "A hybrid system for Parkinson’s disease diagnosis using machine
learning techniques." International Journal of Speech Technology 25.3 (2022): 583-593.

[2] [Link] Haq, Amin, et al. "A survey of deep learning techniques based Parkinson’s disease
recognition methods employing clinical data."Expert Systems with Applications 208 (2022):
118045.

[3] [Link], Adel A., et al. "Metaheuristics with Deep Learning Enabled Parkinson’s
Disease Diagnosis and Classification Model." Journal of Healthcare Engineering 2022
(2022).

[4] [Link], Jing. "Mining imaging and clinical data with machine learning approaches for
the diagnosis and early detection of Parkinson’s disease." npj Parkinson's Disease 8.1 (2022):
1-15.

[5] [Link]š, Máté, et al. "Convolutional neural network ensemble for Parkinson's disease
detection from voice recordings." Computers in biology and medicine 141 (2022): 105021.

[6] Gunjan Pahuja, Bhanu Prasad,Deep learning architectures for Parkinson's disease
detection by using multi-modal features,Computers in Biology and Medicine, Volume146,
2022,105610,ISSN 0010-4825.

[7] Aljalal, M.; Aldosari, S.A.; AlSharabi, K.; Abdurraqeeb, A.M.; Alturki, F.A. Parkinson’s
Disease Detection from Resting-State EEG Signals Using Common Spatial Pattern, Entropy,
and Machine Learning Techniques. Diagnostics 2022, 12, 1033.

[8] Gullapalli, A.S., Mi al, V.K. (2022). Early Detection of Parkinson’s Disease Through
Speech Features and Machine Learning: A Review. In: Senjyu, T., Mahalle, P.N., Perumal, T.,
Joshi, A. (eds) ICT with Intelligent Applications. Smart Innova on, Systems and
Technologies, vol 248. Springer, Singapore.

[9] Kir Raj Bhatele, Anand Jha, Kavish Kapoor, Devanshu Tiwari, Neurodegenerative
diseases-Caps: a capsule network based early screening system for the classification of
neurodegenerative diseases, Cognitive Neurodynamics, 10.1007/s11571-022-09787-1,
(2022).

60
[10] Changqin Quan, Kang Ren, Zhiwei Luo, Zhonglue Chen, Yun Ling,End-to-end deep
learning approach for Parkinson’s disease detection from speech signals, Biocybernetics and
Biomedical Engineering,Volume 42, Issue 2,2022,Pages 556-574,ISSN 0208-5216.

[11] H. Gunduz, "Deep Learning-Based Parkinson’s Disease Classification Using Vocal

Feature Sets," in IEEE Access, vol. 7, pp. 115540-115551, 2019, doi:
10.1109/ACCESS.2019.2936564.

[12] John M. Tracy, Yasin Özkanca, David C. Atkins, Reza Hosseini Ghomi, Investigating
voice as a biomarker: Deep phenotyping methods for early detection of Parkinson's disease.

[13] T. J. Wroge, Y. Özkanca, C. Demiroglu, D. Si, D. C. Atkins and R. H. Ghomi,

"Parkinson’s Disease Diagnosis Using Machine Learning and Voice," 2018 IEEE Signal
Processing in Medicine and Biology Symposium (SPMB), Philadelphia, PA, USA, 2018, pp.
1-7, doi: 10.1109/SPMB.2018.8615607.

[14] Laureano Moro-Velazquez, Jorge A. Gomez-Garcia, Julian D. Arias-Londoño, Najim

Dehak, Juan I. Godino-Llorente, Advances in Parkinson's Disease detection and assessment
using voice and speech: A review of the articulatory and phonatory aspects, Biomedical
Signal Processing and Control, Volume 66, 2021

[15] Turker Tuncer, Sengul Dogan, Udyavara Rajendra Acharya, Automated detec on of
Parkinson's disease using minimum average maximum tree and singular value decomposi on
method with vowels, 2020

[16] Imanne El Maachi, Guillaume-Alexandre Bilodeau, Wassim Bouachir, Deep 1D-

Convnet for accurate Parkinson disease detec on and severity predic on from gait, Expert
Systems with Applica ons, Volume 143, 2020

[17] Gabriel Solana-Lavalle, Juan-Carlos Galán-Hernández, Roberto Rosas-Romero, Automa

c Parkinson disease detec on at early stages as a pre-diagnosis tool by using classifiers and a
small set of vocal features, Biocyberne cs and Biomedical Engineering, Volume 40, Issue 1,
2020

[18] Zehra Karapinar Senturk,Early diagnosis of Parkinson’s disease using machine learning
algorithms, Medical Hypotheses, Volume 138,2020,109603,ISSN 0306-9877.

61
[19] Eduardo Tolosa, Alicia Garrido, Sonja W Scholz, Werner Poewe,Challenges in the
diagnosis of Parkinson's disease,The Lancet Neurology,Volume 20, Issue 5,2021,Pages 385-
397,ISSN 1474-4422.

[20] Parkinson’s disease classifca on using nature inspired feature selec on and recursive
feature elimina on- Prabhleen Kaur Chawla1 ,Meera S. Nair1 Da akumar Gajanan Malkhede1
Hemprasad Yashwant Pa l1· Sumit Kumar Jindal1· Avinash Chandra1 Mahadev Anant
Gawas2- Published -2023.

62