10 VI June 2022

https://doi.org/10.22214/ijraset.2022.44075
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue VI June 2022- Available at www.ijraset.com

Parkinsons Disease Prediction Using Machine

Learning
Mohesh. T1, Gowtham. K2, Vijeesh. P3, Arun Kumar. S4
1, 2, 3, 4
Bachelor of Engineering, Computer Science and Engineering, Anna University

Abstract: Diagnosis of Parkinson's disease (PD) is commonly based on medical observations and assessment of clinical signs,
including the characterization of a variety of motor symptoms. However, traditional diagnostic approaches may suffer from
subjectivity as they rely on the evaluation of movements that are sometimes subtle to human eyes and therefore difficult to
classify, leading to possible misclassification. In the meantime, early non-motor symptoms of PD may be mild and can be caused
by many other conditions. Therefore, these symptoms are often overlooked, making diagnosis of PD at an early stage
challenging. To address these difficulties and to refine the diagnosis and assessment procedures of PD, machine learning
methods have been implemented for the classification of PD and healthy controls or patients with similar clinicalpresentations.

I. INTRODUCTION
Parkinson’s Disease creates neural system disorder for various people. The disease affects the people at different age groups around
the world. Medical research works collaborate with computational intelligence techniques for predicting Parkinson symptoms. PD
has numerous types based on the human abnormalities.
Mostly it disturbs the nature of neural activities and the body movements. Researches evolved in recent years use Machine Learning
(ML) and Deep Learning (DL) approaches for finding early stages of PD. The research works used different types of medical
observations such as voice levels, handwriting variations, body movements, brain signal variations and protein aggregations. These
kinds of observations are measured using various medical apparatuses.

A. Objective
The main objective of this article is to recognize what is Parkinson’s sickness and to discover the early onset of the disorder. We
will use here XGBoost, Random wooded area algorithm, guide Vector Machines (SVMs), and other gadget learning Algorithms
make use of the data-set available on UCL Parkinson Data-set.
https://archieve.ics.uci.edu/ml/datasets/Parkinsonpercent27s+sickness+class.

B. Motivation Of The Project

Parkinson illnesses are the most vital reasons of dying and disability worldwide. According to the Parkinson sickness foundation,
The affected peoples within the global of Parkinson ailment is projected that the 1 million humans are living by way of 2020
in united states of America. The scientific treatment of Parkinson disorder may be recommended on Neuropathologic and
Histopathologic. Medical diagnostic detection of Parkinson sickness can be completed on enormous choice basing on the sensitivity
and specificity of the feature Parkinson disease capabilities consequently, the Parkinson sickness are needed to explore the medical,
pathologic, and nosology research grounded on frequency of prevalence, traits, and including hazard elements of samples.
Parkinson typically affects a huge part of global sufferers over the age of fifty, which has affected to date still now there's no
recognised cause of Parkinson disease, however, it is very likely possible to soothe signs and symptoms knowingly within
the early stage of the subjective patients.
A have a look at claimed that around ninety% of the sufferers affected with vocal damage. The Parkinson remedy is in all
likelihood very expensive.
This reasons most of the sufferers cannot have the funds for the value of the Parkinson sickness. because if the ailment is detected in
the preliminary degree, then the cost will lower and it will additionally be viable to store the patient’s lifestyles these days,
Parkinson disorder prediction is maximum crucial rely for clinical practitioners to take correct decision of such ailment. It’s
a exquisite exercise at gift time, gadget learning based totally giant platform can detect Parkinson ailment.

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 1393
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue VI June 2022- Available at www.ijraset.com

C. Scope Of The Project

In recent, machine learning knowledge of algorithms have generated a great impact and commitment in the Parkinson studies
network for detection of Parkinson disorder. furthermore, system getting to know techniques are distinctive greater precise
outcomes in disease prediction in comparison to others facts taxonomy techniques. prompted by way of this, the authors have
used three distinguished device studying algorithms for detection and right diagnosis of Parkinson sufferers. the main purpose of
this task is to look at the overall performance size of various distinguished class methods for this assignment a couple
of machine studying techniques were used which includes support Vector device, Logistics Regression and different system getting
to know algorithms. moreover, the overall performance of the 3 classifiers become evaluated the use of exceptional methods.

D. Proposed System
By using machine learning techniques, the hassle can be solved with minimum mistakes price. The voice dataset of
Parkinson's ailment from the UCI device mastering library is used as input. additionally our proposed device presents correct results
with the aid of integrating spiral drawing inputs of regular and Parkinson’s affected sufferers. We endorse a hybrid
and accurate results reading affected person both voice and spiral drawing data’s. for that reason combining each the effects,
the medical doctor can finish normality or abnormality and prescribe the medicine based totally at the affected stage.

E. Overview of The Project

Parkinson's disease is a revolutionary nervous system sickness that impacts movement. signs and symptoms begin progressively,
now and again starting with slightly substantive tremor in only one hand. Tremors are not unusual, however he disorder
additionally generally causes stiffness or slowing of movement. within the early degrees of Parkinson's ailment, your face
may display very little expression. Your palms may not swing while you stroll. Your speech may additionally turn out to be
gentle or slurred. Parkinson's sickness signs worsen as your condition progresses through the years. Although Parkinson's
disorder can't be cured, medicines would possibly appreciably enhance your signs and symptoms. occasionally, your doctor may
additionally recommend surgery to regulate certain regions of your brain and enhance your signs and symptoms.

F. Problem Definition
Parkinson's disorder (PD) is a progressive sickness with a presymptomatic interval; this is, there is a length all through which the
pathologic technique has started, but motor symptoms required for the clinical prognosis are absent. in spite of fascinating leads,
direct testing of preclinical markers has been constrained, in particular because there is no reliable manner to identify preclinical
disorder. Idiopathic RBD is characterized via loss of regular atonia with REM sleep. approximately 50% of affected individuals will
increase PD or dementia inside 10 years.

II. LITERATURE REVIEW

Dr. Anupam Bhatia and Raunak Sulekh (2019), “Predictive Model for Parkinson’sDisease through Naive Bayes Classification” In this
study, Naive Bayes was applied to predict the performance of the dataset. Rapid miner 7.6.001 is a tool, which was used to explore,
statistically analyze, and mine the data. The Naive Bayes model performs with 98.5 % accuracy, and 99.75% of precision.
Carlo Ricciardi, et al (2019), “the use of gait analysis’ parameters to categorise Parkinsonism: A information mining method” in
this gadget, Random woodland is used for class in conjunction with comparing it with Gradient
Boosted trees. these consequences are being categorized into three special categories specifically PSP, De Novo
Parkinson’s disorder and stable Parkinson’s disease with their accuracy being as high as 86.4% in comparison to Gradient Boosted
trees which have been correct to a meagre 70%. additionally the precision price of Random forest changed
into most of ninety % against Gradient Boosted trees which were around most of 85%.
Researches have been carried out on PD detection strategies the usage of medical and computational biology strategies. Sadek et al
(2019) designed and carried out artificial Neural network (ANN) and lower back Propagation (BP) techniques for predicting
PD signs and symptoms. ANN and BP techniques have been used for figuring out continuous patient actions. The paintings used
ANN primarily based pattern matching techniques. The patterns of various affected person movements have been registered
and trained for most reliable disorder evaluation. The paintings introduced ideal PD analysis results on homogeneous scientific
functions on the identical time, this paintings were given limitations in multi-variation characteristic evaluation.

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 1394
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue VI June 2022- Available at www.ijraset.com

Gao et al. (2019) proposed a specific prediction and classification technique for Parkinson data analysis. This work was
implemented with the help of data preprocessing techniques, cross fold validations, and ML approaches. The neuro data features
and the tremor data features were analyzed to predict the symptoms. The methods used in this work provided valuable results but
lacked in PD detection and sensitivity rate. Many research works were inherited with different perspectivesfor predicting PD.
In this manner, Mostafa et al. (2020) determined the answer the use of multi-agent facts analysis device. The multi-
agent device turned into designed for evaluating vocal issues. affected person’s vocal statistics were diagnosed as primary
functions for disease detection. The vocal variations were analyzed with the assist of Reinforcement gaining knowledge
of (RL), choice Tree (DT), Naïve Bayes category and Random forest (RF) techniques. This paintings gathered the medical facts
from Tel Aviv Sourasky medical middle however, the paintings became lagging with real-time problems. Biswajt et al. (2020)
proposed the system to locate D symptoms the usage of speech attributes. on this work, SVM and Naïve Byes strategies have
been mentioned the use of voice information features. the radical paintings built ML primarily based voice evaluation models
for locating the PD symptoms. The work tried to task extra correct outcomes. however, this work lacked with the constrained dataset
features.
Rastegari et al (2021) evolved data gain evaluation model for detecting PD capabilities from the given dataset. This technique
used diverse ML and data gain approached in mixed manner. This method labored well in PD findings. however they produced
insignificant effects as compared to DL primarily based PD evaluation models. Seppi et al. (2021) analyzed and carried
out Parkinson treatment evaluation for various non-motor signs. This work cautioned that the remedy analysis allows to improve
and update the future subsequent level treatments. on this regard, the paintings gathered the evidences of numerous treatments and
produced valuable suggestions. Many systems created clinical review reviews on PD and remedies however those structures were
no longer equipped with own specific techniques.
Espay et al. (2022) proposed new strategies for comparing Synuclein protein disorders and aggregation price for the detection of
Parkinson and Alzheimer signs. specifically, the protein aggregation techniques used for sporadic Parkinson and Alzheimer. At that
point, the gadget did not take DL and ML help to educate the protein capabilities. locating one of a kind styles of PD and predicting
the signs and symptoms were virtually complicated tasks.
Bot et al. (2022) used cell datasets and mobile primarily based statistics collections for PD evaluation. on this work, the techniques
have been educated to supply less overhead statistics functions on the detection of Parkinson. The techniques were correctly
operated but restricted with scalability issues.
Bouwmans et al. (2022) passed through with Idiopathic PD (IPD) symptoms for assorted feature class Ramzi M. Sadek et al (2022),
“Parkinson’s disease Prediction the use of artificial Neural network” on this machine, 195 samples in the dataset had been divided
into one hundred seventy training samples and 25 validating samples. Then importing the dataset inside the just Neural network
(JNN) environment, we skilled, established the synthetic Neural community model. The maximum crucial attributes contributing to
the ANN model were made known of. The ANN version changed into 90% correct
Satish Srinivasan, Michael Martin & Abhishek Tripathi (2020), “ANN based totally information Mining analysis of Parkinson’s
disorder” on this study, it was supposed to understand how the exceptional types of preprocessing steps should affect the prediction
accuracy of the classifier. in the method of classifying the Parkinson’s disorder dataset the use of the ANN based MLP classifier a
appreciably high prediction accuracy changed into discovered while the dataset turned into pre-processed the usage of each the
Discretization and Resample approach, both inside the case of 10-fold go validation and 80:20 split.

III. SOFTWARE REQUIRMENTS SPECIFICATION

A. Python
1) Python 3

B. Libraries
1) Numpy
2) pandas
3) joblib
4) sklearn
5) argparse
6) pickle
7) matplotlib, etc.

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 1395
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue VI June 2022- Available at www.ijraset.com

C. Operating System
1) Windows or Ubuntu

D. Hardware Requirements Specification

1) Laptop with basic hardware
2) Webcam

E. Inputs
1) pd_speech_features.csv
IV. SYSTEM ANALYSIS
A. System Review
This survey is executed to recognize the need and prerequisite of the general population who has PD, and to do as such, we went
through exceptional sites and programs and searched for the essential statistics. based on these information, we made an audit that
helped us get new thoughts and make distinctive arrangements for our challenge. We reached the choice that there's a need of such
utility and felt that there's a respectable extent of development on this field too.

B. Technology Used
1) Python: Python is an interpreted, high-level, general-purpose programming language. Python's design philosophy emphasizes
code readability with its notable use of significant whitespace. Its language constructs and object- oriented approach aim to help
programmers write clear, logical code for small and large-scale projects. Python is dynamically typed AND supports multiple
programming paradigms, including procedural, object-oriented, and functional programming.
2) Pycharm: Pycharm makes it easier for programmers to write various web applications in python supporting widely used web
technologies like HTML, CSS, JavaScript and CoffeeScript. It even simplifies isomorphic web application development by
supporting both AngularJS and NodeJS
3) machine learning: Machine learning is the scientific study of algorithms and statistical models that computer systems use in
order to perform a specific task effectively without using explicit instructions, relying on patterns andinference instead. It is seen
as a subset of artificial intelligence.
4) Machine Learning Datasets: Machine learning has been developed for decades, and therefore there are some datasets of
historical significance. One of the most well-known repositories for these datasets is the UCIMachine Learning Repository. Most
of the datasets over there are small in size because the technology at the time was not advanced enough to handle larger size
data. Some famous datasets located in this repository are the iris flower dataset
V. REQUIREMENT ANALYSIS
A. Python
Python is the basis of the program that we wrote. It utilizes many of thepython libraries.

B. Libraries
1) NumPy: NumPy, which stands for Numerical python, is a library consistingof multidimensional array objects and a collection of
routines for processing those arrays. Using NumPy mathematical and logical operations on arrays can be performed. NumPy is
a python Pre-requisite for Dlib.
2) Pandas: Pandas is an open source Python package that is most widely used for data science/data analysis and machine learning
tasks. It is built on top of another package named Numpy, which provides support for multi-dimensional arrays.
3) Matplotlib: Matplotlib is a cross-platform, data visualization and graphical plotting library for Python and its numerical
extension NumPy. As such, it offers aviable open source alternative to MATLAB.
4) Sklearn: Scikit-learn is a free machine learning library for Python. It features various algorithms like support vector machine,
random forests, and k-neighbours,and it also supports Python numerical and scientific libraries like NumPy and SciPy.
5) Argparse: The argparse module in Python helps create a program in a command-line-environment not only easy to code but
also improves interaction. The argparse module also automatically generates help and usage messages and issues errors when
users give the program invalid arguments.
6) Joblib: Joblib is a set of tools to provide lightweight pipelining in Python. In particular: transparent disk-caching of functions
and lazy re-evaluation (memorize pattern) easy simple parallel computing.

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 1396
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue VI June 2022- Available at www.ijraset.com

C. OS
Program is tested on Windows 10 build 1903 and PopOS 19.04

D. Classification Techniques
1) Logistics Regression: Logistic Regression changed into commonly used in the organic studies and packages in the early
twentieth century. Logistic Regression (LR) is one of the maximum used device learning algorithms that is used wherein
the goal variable is categorical. lately, LR is a famous method for binary classification troubles. moreover, it presents a discrete
binary product between zero and 1. Logistic Regression computes the relationship among the feature variables by means of
assessing possibilities (p) the use of underlying logistic function.
2) Support Vector Machine (SVM): guide vector system has been first added via Vladimir Vapnik and Alexey Chervonenkis. SVM
is a way of device studying that can clear up both linear and nonlinear issues. It presents exact overall performance to remedy
each regression and classification hassle. The SVM classification technique inspects for the highest quality separable
hyperplane if you want to classify the dataset between two instructions. eventually, the model can estimate noisy information
troubles for brand new instances
3) Ada Boost: AdaBoost combines the predictions from brief one-degree decision timber, referred to as selection stumps, despite
the fact that other algorithms also can be used. selection stump algorithms are used as the AdaBoost set of rules seeks to apply
many susceptible fashions and correct their predictions with the aid of including additional weak models. The education
algorithm includes beginning with one choice tree, locating those examples within the training dataset that have been
misclassified, and including more weight to the ones examples. another tree is educated at the same records, despite the fact that
now weighted with the aid of the misclassification mistakes. This technique is repeated until a preferred variety of trees are
introduced
4) Gradient Boost: Unlike, Adaboosting set of rules, the base estimator within the gradient boosting algorithm can
not be cited through us. the bottom estimator for the Gradient raise algorithm is fixed and i.e. decision Stump. Like,
AdaBoost, we will music the n_estimator of the gradient boosting set of rules. but, if we do not point out the fee of n_estimator,
the default price of n_estimator for this algorithm is 100.Gradient boosting set of rules can be used for predicting not
simplest non-stop goal variable (as a Regressor) but also specific target variable (as a Classifier). while it's far used as a
regressor, the value characteristic is suggest square errors (MSE) and while it's far used as a classifier then the cost function is
Log loss
5) Random Forest: Random woodland is a Supervised device mastering algorithm this is used widely in classification and
Regression troubles. It builds choice bushes on extraordinary samples and takes their majority vote for type and average in case
of regression. one of the most vital functions of the Random woodland set of rules is that it can manage the information set
containing continuous variables as within the case of regression and specific variables as within the case of category. It plays
higher outcomes for category issues
6) Naive Bayes: Naive Bayes algorithm is a supervised learning algorithm, which is based on Bayes theorem and used for solving
classification problems. It is mainly used in text classification that includes a high-dimensional training dataset. Naïve Bayes
Classifier is one of the simple and most effective Classification algorithms which helps in building the fast machine learning
models that can make quick predictions. It is a probabilistic classifier, which means it predicts on the basis of the probability of an
object. Some popular examples of Naïve Bayes Algorithm are spam filtration, Sentimental analysis, and classifying articles.
7) Neural Network: Neural networks, also referred to as artificial neural networks (ANNs) or simulated neural networks (SNNs),
are a subset of gadget gaining knowledge of and are on the heart of deep learning algorithms. Their call and structure are
stimulated via the human brain, mimicking the way that organic neurons sign to one another. synthetic neural networks (ANNs)
are constructed from a node layers, containing an input layer, one or extra hidden layers, and an output layer. every node, or
synthetic neuron, connects to another and has an associated weight and threshold. If the output of any individual node is above
the required threshold value, that node is activated, sending facts to the next layer of the community. in any other case, no
records is exceeded alongside to the subsequent layer of the community
8) XGBoost: XGBoost, which stands for severe Gradient Boosting, is a scalable, disbursed gradient-boosted choice tree (GBDT)
device mastering library. It gives parallel tree boosting and is the leading machine mastering library for regression, class, and
ranking issues. It’s important to an understanding of XGBoost to first draw close the system getting to know standards and
algorithms that XGBoost builds upon: supervised device mastering, choice bushes, ensemble getting to know, and grading
raise. Supervised machine learning uses algorithms to educate a version to find patterns in a dataset with labels and functions

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 1397
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue VI June 2022- Available at www.ijraset.com

and then uses the educated model to are expecting the labels on a brand new dataset’s functions.
9) Decision Tree: The decision of making strategic splits heavily impacts a tree’s accuracy. The choice criteria are exceptional for
type and regression timber. decision timber use a couple of algorithms to decide to break up a node into or greater sub-nodes.
The introduction of sub-nodes will increase the homogeneity of resultant sub- nodes. In other phrases, we will say that the
purity of the node increases with admire to the target variable. The decision tree splits the nodes on all available variables after
which selects the break up which results in most homogeneous sub-nodes.

E. Laptop
Used to run our code.

VI. SYSTEM DESIGN

A. Architectural Diagram

FIG 6.1 Architectural Diagram

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 1398
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue VI June 2022- Available at www.ijraset.com

VII. IMPLEMENTATION
A. Parkinson's Disease Prediction

FIG 7.1 Parkinson Disease

B. Running Training Models

FIG 7.2 Pre-Training Model

C. See run_model.py
1) Use python run_model.py --help to see the complete list of trained models.
2) Choose a model with the python run_model.py --model {model_number} to use thepretrained model on the parkinsons's disease
dataset.
TO RUN
 EDA.py to see the data analysis visual plots.
 generate_weights.py to save weights for models with best hyperparameters.
 run_model.py to see results on the saved weights.

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 1399
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue VI June 2022- Available at www.ijraset.com

FIG 7.3 Traning ANG Testing

D. Generating Weights
See generate_weights.py
1) Use python generate_weights.py --help to see the complete list of trained models.
2) Choose a model with the python generate_weights.py --model {model_number} togenerate fresh set of weights from
best_params.py.

Name Description
Run Model from pre-trainedweights
run_model.py
Get Models for besthyperparameters
best_params.py
generate_weights.py Save Weights for best models
PCA, Feature Selection, train-testsplit
pre_processing.py
Exploratory Data Analysis ondataset
EDA.py
weights_original Weights from the Original Project

VIII. PROGRAM AND SCREENSHOTS OF PROJECT

A. best_params.py
from sklearn.naive_bayes import BernoulliNB
from sklearn.linear_model import LogisticRegression
from sklearn.ensemble import RandomForestClassifier, GradientBoostingClassifier
from sklearn.tree import DecisionTreeClassifier
from xgboost import XGBClassifier
from sklearn.svm import SVC
from sklearn.ensemble import AdaBoostClassifier
from sklearn.model_selection import GridSearchCV
from pre_processing import X_train, X_test, y_test, y_train
from sklearn.neural_network import MLPClassifier

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 1400
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue VI June 2022- Available at www.ijraset.com

# best_params file returns the best_models tuned to best performing parameters

best_models = {1: "naive_bayes", 2: "logistic_regression", 3: "decision_tree",

4: "svm", 5: "random_forest", 6: "ada_boost", 7: "gradient_boost", 8: "xg_boost",
9: "neural_net"}

def best_model(model_num):
# Naive Bayes
if model_num == 1:
clf = BernoulliNB()
clf.fit(X_train, y_train)
return clf

# Logistic Regression
if model_num == 2:
param_grid = {'penalty': ['l1', 'l2'],
'C': [0.001, 0.01, 0.1, 1, 10, 100, 1000]}
grid = GridSearchCV(LogisticRegression(random_state=0),
param_grid, refit=True)
grid.fit(X_train, y_train)
return grid

# decision_tree

if model_num == 3:
param_grid = {'criterion': ['gini', "entropy"],
'splitter': ["best", "random"],
'max_depth': range(1, 20),
'min_samples_split': range(2, 10, 1)}
grid = GridSearchCV(DecisionTreeClassifier(random_state=0),
param_grid, refit=True)
grid.fit(X_train, y_train)
return grid

# SVM
if model_num == 4:
param_grid = {'C': [0.1, 1, 10, 100, 1000],
'gamma': [1, 0.1, 0.01, 0.001],
'kernel': ['rbf', 'poly', 'sigmoid']
}
grid = GridSearchCV(SVC(random_state=0), param_grid, refit=True)
grid.fit(X_train, y_train)
return grid

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 1401
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue VI June 2022- Available at www.ijraset.com

# Random Forest
if model_num == 5:
param_grid = {'n_estimators': range(10, 100, 5),
'criterion': ['gini', 'entropy'],
'max_depth': range(1, 20),
'min_samples_split': range(2, 10, 1),
'max_features': ['auto', 'sqrt', 'log2']
}
grid = GridSearchCV(RandomForestClassifier(random_state=0, n_jobs=10),
param_grid, refit=True, n_jobs=10)
grid.fit(X_train, y_train)
return grid

# ADA Boost
if model_num == 6:
param_grid = {'n_estimators': range(10, 100, 5),
'algorithm': ['SAMME', 'SAMME.R'],
'learning_rate': [0.001, 0.01, 0.1, 1.0, 10.0]
}
grid = GridSearchCV(AdaBoostClassifier(random_state=0),
param_grid, refit=True)
grid.fit(X_train, y_train)
return grid

# Gradient Boost
if model_num == 7:
param_grid = {'n_estimators': [100, 150, 200, 250, 300, 350, 400, 450, 500],
'criterion': ['friedman_mse', 'mse'],
'max_features': ['sqrt', 'log2'],
'learning_rate': [0.001, 0.01, 0.1, 1.0, 2, 4],
'loss': ['deviance', 'exponential'],
}
grid = GridSearchCV(GradientBoostingClassifier(
random_state=0), param_grid, refit=True)
grid.fit(X_train, y_train)
return grid
# XG Boost
if model_num == 8:
param_grid = {'n_estimators': range(0, 100, 5),
'learning_rate': [0.01, 0.1],
'max_depth': range(2, 6)
}

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 1402
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue VI June 2022- Available at www.ijraset.com

grid = GridSearchCV(XGBClassifier(random_state=0),
param_grid, refit=True)
grid.fit(X_train, y_train)
return grid

# Neural Net
if model_num == 9:
param_grid = {'alpha': [x * 0.0001 for x in range(1, 1000, 2)],
'learning_rate_init': [0.01, 0.001, 0.0001],
'max_iter': [100, 300, 1000],
'hidden_layer_sizes': [(128, 64, 32), (128, 32, 8), (256, 128, 64, 32), (64,
32, 16), (64, 32), (64, 16)],
}
grid = GridSearchCV(MLPClassifier(random_state=42),
param_grid, refit=True, n_jobs=10)
grid.fit(X_train, y_train)
return grid
B. generate_weights.py

import argparse
import pickle
import sys
from best_params import best_model

if name == " main ":

# Adding Arguments to make user select one model to run results on
parser = argparse.ArgumentParser(
formatter_class=argparse.RawTextHelpFormatter)
model_dict = {1: "naive_bayes", 2: "logistic_regression", 3: "decision_tree",
4: "svm", 5: "random_forest", 6: "ada_boost", 7: "gradient_boost", 8: "xg_boost",
9: "neural_net", 10: "all"}
parser.add_argument(
"-m", "--model", type=int, choices=list(range(1, len(model_dict)+1)), help="\
\n1. Naive Bayes \
\n2. Logistic Regression \
\n3. Decision Tree\
\n4. SVM\
\n5. Random Forest\
\n6. AdaBoost\
\n7. Gradient Boost\
\n8. XGBoost\
\n9. Neural Network\
\n10. All Models\
")
args = parser.parse_args()

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 1403
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue VI June 2022- Available at www.ijraset.com

if not args.model:
print("Model not provided. See run_model.py --help")
sys.exit()
def save_weights(model_num):
model = best_model(model_num)
filename = f"./weights/{model_dict[model_num]}.sav"
print(f"\nSaving {filename} ...")
pickle.dump(model, open(filename, 'wb'))
# save the model to disk
if(model_dict[args.model] != "all"):
save_weights(args.model)
else:
# Run for all models
for model_num in range(1, len(model_dict)):
save_weights(model_num)
else:
print("Run the script run_model.py to obtain results.")

C. pre_processing.py

import pandas as pd
import numpy as np
from sklearn.feature_selection import f_classif, SelectKBest
from sklearn.preprocessing import StandardScaler
from sklearn.decomposition import PCA

df = pd.read_csv('./dataset/pd_speech_features.csv')
id = df['id']
label = df['class']
n = len(id)
id_label = {}
for i in range(n):
id_label[id[i]] = label[i]

train_healthy_id = []
test_healthy_id = []
train_pd_id = []
test_pd_id = []
train_n_healthy = 58
test_n_healthy = 6
train_n_pd = 170
test_n_pd = 18

train_count_healthy = 0

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 1404
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue VI June 2022- Available at www.ijraset.com

test_count_healthy = 0
train_count_pd = 0
test_count_pd = 0
for key, val in id_label.items():
if val == 0:
if train_count_healthy < train_n_healthy:
train_healthy_id.append(key)
train_count_healthy += 1
else:
test_healthy_id.append(key)
train_count_healthy += 1
else:
if train_count_pd < train_n_pd:
train_pd_id.append(key)
train_count_pd += 1
else:
test_pd_id.append(key)
test_count_pd += 1

train_healthy_id.extend(train_pd_id)
test_healthy_id.extend(test_pd_id)
train_id = train_healthy_id
test_id = test_healthy_id
train_id = np.array(train_id)
test_id = np.array(test_id)

trainX = None
trainY = None
for i in range(train_id.shape[0]):
vals = df[df['id'] == train_id[i]].values
if i == 0:
trainX = vals[:, 1:-1]
trainY = vals[:, -1].reshape((-1, 1))
else:
trainX = np.append(trainX, vals[:, 1:-1], axis=0)
trainY = np.append(trainY, vals[:, -1].reshape((-1, 1)), axis=0)

testX = None
testY = None
for i in range(test_id.shape[0]):
vals = df[df['id'] == test_id[i]].values
if i == 0:
testX = vals[:, 1:-1]

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 1405
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue VI June 2022- Available at www.ijraset.com

testY = vals[:, -1].reshape((-1, 1))

else:
testX = np.append(testX, vals[:, 1:-1], axis=0)
testY = np.append(testY, vals[:, -1].reshape((-1, 1)), axis=0)
trainY = trainY.reshape((-1,))

trainX_copy = trainX
trainY_copy = trainY
testX_copy = testX
testY_copy = testY

sel = SelectKBest(f_classif, k=490)

trainX_copy = sel.fit_transform(trainX_copy, trainY_copy)
testX_copy = sel.transform(testX_copy)

sc = StandardScaler()
trainX_copy = sc.fit_transform(trainX_copy)
testX_copy = sc.transform(testX_copy)

pca = PCA(n_components=0.99)
trainX_copy = pca.fit_transform(trainX_copy)
testX_copy = pca.transform(testX_copy)

if name == " main ":

print("Data Pre-processed Succesfully. Run the script run_model.py")
else:
X_train, X_test, y_test, y_train = trainX_copy, testX_copy, testY_copy, trainY_copy

D. run_model.py
import argparse
import pickle
import Joblib
import sys
# Loading the dataset
from pre_processing import X_train, X_test, y_test, y_train

# Adding Arguments to make user select one model to run results on

parser = argparse.ArgumentParser(formatter_class=argparse.RawTextHelpFormatter)
model_dict = {1: "naive_bayes", 2: "logistic_regression", 3: "decision_tree",
4: "svm", 5: "random_forest", 6: "ada_boost", 7: "gradient_boost", 8: "xg_boost", 9:
"neural_net", 10: "all"}

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 1406
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue VI June 2022- Available at www.ijraset.com

parser.add_argument(
"-m", "--model", type=int, choices=list(range(1, len(model_dict)+1)), help="\
\n1. Naive Bayes \
\n2. Logistic Regression \
\n3. Decision Tree\
\n4. SVM\
\n5. Random Forest\
\n6. AdaBoost\
\n7. Gradient Boost\
\n8. XGBoost\
\n9. Neural Network\
\n10. All Models\
")

args = parser.parse_args()
if not args.model:
print("Model not provided. See run_model.py --help")
sys.exit()

# Loading and Running the Model

def load_and_predict(model_num):
filename = f"./weights/{model_dict[model_num]}.sav"
print(f"\nLoading {filename} ...")

loaded_model = pickle.load(open(filename, 'rb'))

result = loaded_model.score(X_test, y_test)

print(f"({model_dict[model_num]}) Model Accuracy: {result}%")

if(model_dict[args.model] != "all"):
load_and_predict(args.model)
else:
# Run for all models
for model_num in range(1, len(model_dict)):
load_and_predict(model_num)

E. EDA.py
from pre_processing import X_train, X_test, y_test, y_train
import pandas as pd
import numpy as np

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 1407
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue VI June 2022- Available at www.ijraset.com

import seaborn as sns

import matplotlib.pyplot as plt
from sklearn.manifold import TSNE

# Perform EDA
df = pd.read_csv('./dataset/pd_speech_features.csv')

print('Dataset Shape:')
print(df.shape)
print()

print('Class Distribution:')
print(df['class'].value_counts())
print()

positives = df[df['class'] == 1]
negatives = df[df['class'] == 0]

print('Positive Class (Patients with PD) Gender Distribution')

print(positives['gender'].value_counts())
print()

print('Negative Class (Patients without PD) Gender Distribution')

print(negatives['gender'].value_counts())
print()

# Number of patients counts

def print_patient_dist(df):
ids = np.unique(df['id'].values)
fem = 0
male = 0
for i in ids:
if df[df['id'] == i]['gender'].iloc[0] == 0:
fem += 1
else:
male += 1
print('Unique Males: ', male)

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 1408
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue VI June 2022- Available at www.ijraset.com

print('Unique Females: ', fem)

print('Positive Class Unique Patients Distribution')

print_patient_dist(positives)
print()
print('Negative Class Unique Patients Distribution')
print_patient_dist(negatives)
print()

print('Database Info')
print(df.info())
print()

# Plot distribution of 3 important features

df_selected = df[['DFA', 'mean_MFCC_2nd_coef', 'tqwt_entropy_log_dec_12']]

print(df_selected.describe())

plt.gcf().set_size_inches(12, 5)
ax = sns.histplot(df['DFA'], color='#3EADA7', alpha=1, edgecolor='white')
ax.set_title('DCA')
plt.show()

plt.gcf().set_size_inches(12, 5)
ax = sns.histplot(df['mean_MFCC_2nd_coef'],
color='#3EADA7', alpha=1, edgecolor='white')
ax.set_title('mean_MFCC_2nd_coef')
plt.show()

plt.gcf().set_size_inches(12, 5)
ax = sns.histplot(df['tqwt_entropy_log_dec_12'],
color='#3EADA7', alpha=1, edgecolor='white')
ax.set_title('tqwt_entropy_log_dec_12')
plt.show()

# Peform tsne visualization on the preprocessed data

tsne_results = TSNE(n_components=2, random_state=1).fit_transform(X_train)

DF = pd.DataFrame({'y': y_train.reshape((-1,))})

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 1409
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue VI June 2022- Available at www.ijraset.com

DF['tsne1'] = tsne_results[:, 0]
DF['tsne2'] = tsne_results[:, 1]

plt.figure(figsize=(5, 5))
ax = sns.scatterplot(
x="tsne1", y="tsne2",
hue="y",
palette=sns.color_palette("hls", 2),
data=DF,
legend=False,
alpha=1
)

ax.set_title('TSNE Visualization')
plt.show()

F. Screenshots

Fig 8.1 DCA

Fig 8.2 TSNE Visualization

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 1410
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538
Volume 10 Issue VI June 2022- Available at www.ijraset.com

IX. CONCLUSION
Parkinson’s Disease is a totally grave disease and has no cure till date. since it impacts the actions of the parts of the body, the
speech additionally stands affected. here, the gadget tries to offer a way of detecting Parkinson’s ailment so one can bring about a
quick action to reduce or even put off it from affecting the whole body. This gadget aims to make this method of expertise a case of
Parkinson’s on the earliest via each, the affected person as well as scientific experts. hence, the goal is to apply numerous machine
getting to know strategies like SVM, choice Tree, for buying the maximum accurate result. Here using Decision Tree and building a
classifier results in an accuracy of 94.7774555%.

REFERENCES
[1] Adrien Payan, Giovanni Montana, Predicting Alzheimer's disease: a neuroimaging study with 3D convolutional neural networks.
[2] Alemami, Y. and Almazaydeh, L. (2020) Detecting of ParkinsonDisease through Voice Signal Features. Journal of American Science,
[3] Fayao Liu, Chunhua Shen, Learning Deep Convolutional Features for MRI Based Alzheimer’s Disease Classification.
[4] Hadjahamadi, A.H. and Askari, T.J. (2020) A Detection Support System for Parkinson’s Disease Diagnosis Using Classification and Regression Tree. Journal of
Mathematics and Computer Science , 4, 257-263.
[5] Little, M.A., McSharry, P.E., Hunter, E.J. and Ramig, L.O. (2018), Suitability of Dysphonia Measurements for Telemonitoring of Parkinson’s disease. IEEE
Transactions on Biomed ical Engineering, 56, 1015-1022.
[6] Muhlenbach, F. and Rakotomalala, R. (2019) Discretization of Continuous Attributes. In: Wang, J., Ed., Encyclopedia of Data Warehousing and Mining, Idea
Group Reference, 397-402.
[7] A. J. Espay, ‘‘Revisiting protein aggregation as pathogenic in sporadicparkinson and alzheimer diseases,’’ Neurology, vol. 92, no. 7, pp. 329–337,Feb. 2019.
[8] B. M. Bot, C. Suver, E. C. Neto, M. Kellen, A. Klein, C. Bare, M. Doerr, A. Pratap, J. Wilbanks, E. R. Dorsey, S. H. Friend, and A. D. Trister, ‘‘The mPower
study, parkinson disease mobile data collected using ResearchKit,’’ Sci. Data, vol. 3, no. 1, Dec. 2019, Art. no. 160011.
[9] A. E. P. Bouwmans, A. M. M. Vlaar, W. H. Mess, A. Kessels, and W.
[10] E. J. Weber, ‘‘Specificity and sensitivity of transcranial sonography of the substantia nigra in the diagnosis of Parkinson’s disease: Prospective cohort study in
196 patients,’’ BMJ Open, vol. 3, no. 4, 2020, Art. no. e002613.
[11] Pickle, N. T., Shearin, S. M., and Fey, N. P. Dynamic neural network approach to targeted balance assessment of individuals with and without neurological
disease during non-steady-state locomotion. Journal of Neuroengineering and Rehabilitation 16 (JUL 12 2020), 88. PT: J; NR: 31; TC: 0; J9: J NEUROENG
REHABIL; PG: 9; GA: IJ0RY; UT: WOS:000475608600003.

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 | 1411