PARKINSON DISEASE DETECTION

USING MACHINE LEARNING

Dr.M.Sangeetha [1] K.Muthamizhmaran[2] C.Marianus Prince[3]

Professor UG Scholar UG Scholar
Department of CSE Department of CSE Department of CSE
Panimalar Engineering college Panimalar Engineering College Panimalar Engineering College
Poonamallee, Chennai. Poonamallee, Chennai. Poonamallee, Chennai.

Abstract of the improved technology, it is now possible to

analyze the voice of an individual to tell whether that
Parkinson's Disease (PD) is a progressive neurological person is suffering from Parkinson’s Disease. Voice
condition that impacts movement, marked by samples can be analyzed for various acoustic measures
symptoms such as tremors, rigidity, and challenges such as jitter and shimmer, HNR and other phonatory
with balance and coordination. It is equally crucial to measures obtained from the speech samples. It is well
detect PD early for better management of the disease established that using these features when training
and improving the patients’ quality of life. machine learning models for Parkinson disease allows
Nevertheless, traditional diagnostic methods are them to achieve very high accuracy in the classification
primarily based on clinical observation and may not of sick and healthy individuals. Such a methodology is
be very effective in the detection of early stages of the promising regarding the screening and early detection of
disease.The research employs a dataset of patient the disease as it is non-invasive.
voice recordings with a focus on voice dysfunction,
which is one of the main features of PD. In order to This approach has the potential to be beneficial in the
create a classificatory model, we process voice field, especially due to its use in distant healthcare
recordings and include parameters such as jitter, services where it is cheap and therefore enhances early
shimmer, and the harmonics to noise ratio in our detection and follow up of this disease which in turn helps
processed sound samples. We analyze several improve patient care and management. Through the use
machine learning methods – Support Vector Machines of sophisticated machine learning algorithms, such
SVM, Random Forest and Neural Networks in our speech pattern recognition system can be modeled and
case - to classify subjects with and without realized with a high degree of accuracy.
Parkinson's disease. The performance of trained
models is expressed using accuracy, precision, recall, Text to Speech Software and m-Health Services are also
F1-score and the models are trained on the features helpful Rehabilitative Technological Systems. Due to
obtained at the extraction phase. the rapid advancement of smart technologies, machine
learning systems can identify some vocal markers of
Parkinson’s disease. Such systems can apprehend the
Introduction characteristics of speech which might not be
distinguishable to the ears of a human being, thus
Parkinson's Disease (PD) is a persistent condition permitting timely assistance and customization of
marked by degeneration of the nervous system, which management techniques.
primarily affects the motor ability and leads to
tremors, stiffness and trouble in performing even basic
physical tasks. Voice related psychopathology may
develop prior to the onset of discernable tremors, in
the form of changes such as lowered volumes, flat
intonations and the absence of rate variations of
speech. Such speech changes allow identifying
patients at risk of developing Parkinson's disease at an
early stage, which is crucial for ensuring effective
therapy and control of the disease. Parkinson’s Speech
Analysis using Machine Learning (ML) Algorithms In
the recent past, because
Related Work In [7], Meireles and Massano (2012) reviewed cognitive
impairment and dementia in Parkinson's disease,
In [1], Max A. Little et al. (2008) described a novel
examining the clinical features, diagnostic criteria, and
method of telemonitoring Parkinson’s disease by
management strategies. Their work emphasized the need
means of dysphonia analysis.The strategy utilized
for tailored approaches to cognitive symptom
Bolster Vector Machines (SVMs) to classify patients
management in PD patients to improve outcomes.
based on extricated voice highlights, accomplishing a
exceptional 91 accuracy in recognizing between PD
patients and sound subjects. This work grandstands Samii, Nutt, and Ransom [8] (2004) provided a
the potential of wearing gadgets with implanted voice comprehensive overview of Parkinson's disease,
empowering machine learning methods for farther focusing on its epidemiology, pathophysiology, clinical
persistent observing. features, and treatment. Their review underscored the
importance of early diagnosis and a multidisciplinary
approach to manage the disease and improve the quality
of life for patients.
In [2], Sankar et al. (2018) proposed the use of the
Tunable Q-factor Wavelet Transform (TQWT) to In [9], R. Das (2010) compared different machine
enhance Parkinson's disease telemonitoring through learning- based classification strategies for diagnosing
vocal analysis. The framework combined TQWT with Parkinson's infection. The consider assessed the
Mel- recurrence cepstral coefficients (MFCCs) and adequacy of a few calculations,highlighting their
analyzed voice information from 252 members, potential for improving early detection and aiding in
appearing progressed classification exactness for clinical decision-making for PD management.
farther PD determination compared to conventional
strategies. Finally, in [10], Litvan et al. (2012) developed
diagnostic criteria for mild cognitive impairment in
In [3], Pereira et al. (2017) leveraged Convolutional Parkinson's disease through guidelines set by the
Neural Networks (CNNs) to enhance PD detection by Movement Disorder Society Task Force. Their work
analyzing diverse health data, including motor contributed to the early identification of cognitive issues
activities and sensor inputs. Their system effectively in PD, supporting better patient management and
identified Parkinson's disease, demonstrating CNN's tailored treatment strategies.
potential for fast and accurate automatic diagnosis in
clinical applications.
Existing System
In [4], Vallejo et al. explored multi-objective
regression models to predict motor and cognitive The current methods for detecting Parkinson’s Disease
deficits in Parkinson's disease. By integrating multiple (PD) through machine learning have mainly
health indicators, their approach improved diagnostic concentrated on creating predictive models that evaluate
accuracy and provided deeper insights into the particular data inputs, including voice recordings, motor
variability of PD symptoms, helping to facilitate early skills, and handwriting styles. These systems are
diagnosis and personalized patient care. designed to enhance or potentially surpass conventional
diagnostic approaches by offering a non-invasive,
In [5], Lones et al. (2017) developed a home affordable, and early detection option for Parkinson’s
monitoring system based on evolutionary algorithms Disease. Below is a summary of the key features of the
to assess dyskinesia in Parkinson's disease patients. existing machine learning systems utilized for PD
The system enables real-time remote monitoring and detection using Python.
provides valuable insights into symptom fluctuations,
improving patient care through continuous data
collection and analysis. Proposed System
In [6], Pereira et al. (2015), in their work, looked into The suggested system is designed to improve the early
the feasibility of automated diagnosis of Parkinson’s identification of Parkinson’s Disease (PD) through the
disease through handwriting analysis. The study use of machine learning techniques developed in
assessed the threat potential of handwriting Python. By concentrating on the examination of voice
movements as an adjuvant aid to Parkinson’s disease data, which is recognized for indicating early symptoms
diagnosis that would be non-invasive, and thus of PD like vocal difficulties, this system aims to
preferable over existing methods. establish a non- invasive, efficient, and affordable
approach for detecting individuals at risk of
Parkinson’s Disease. The primary
goal is to create a machine learning model that can
such as vocal characteristics (MDVP, jitter, shine, etc.)
effectively differentiate between those with
that are valuable for distinguishing the illness. The
Parkinson’s Disease and healthy individuals, thus
information is at that point handled, where criteria for
enhancing traditional clinical diagnostic practices.
PD distinguishing proof and classification are
extricated. After designing the information into a
The initial stage entails the collection of audio or appropriate structure, it is utilized to prepare the SVM
voice data for that matter which could either be drawn demonstrate.
from an open source collection or recorded newly.
This information is then preprocessed which entails The demonstrate goes through an iterative preparing
cleaning the data, for instance removing irreverent handle, where it minimizes classification blunders. In
columns, replacing or filling any empty cells and case the blunder rate is over an satisfactory edge, the
standardizing the data across every dimension. When show is retrained. Once the preparing accomplishes
the feature extraction stage is complete, the next stage palatable exactness, the SVM demonstrate is recreated
involves the implementation of certain machine or adjusted to classify people based on their voice
learning methods like Support Vector Machine highlights. The comes about are at that point passed
(SVM) or Random Forest to build a model that is through a classifier that categorizes the information as
capable of classifying individuals. This is done using either "Typical" (showing the nonappearance of
already labeled data, which enables the model to learn Parkinson's) or "PD" (showing the nearness of
and identify the differences between people who have Parkinson's Malady).Once the SVM show accomplishes
Parkinson’s disease and those who do not. Amid the the required precision and the blunder rate is
preparing organize, the calculation is able to minimized, it is then utilized for classification within
memorize the highlights of Parkinson's illness in the genuine world. The classifier takes in unused voice
modern voice recordings and claim whether there's information and applies the prepared demonstrate to
Parkinson's malady display within the modern voices anticipate whether the person has Parkinson's Malady
or not depending upon the highlights extricated. (spoken to by 'PD') or is sound (spoken to by 'Normal').
This decision- making step is pivotal for the framework
Architecture Diagram of to supply precise symptomatic help in recognizing PD.

PDD In general, this flowchart speaks to a exceedingly

efficient approach to the discovery of Parkinson's
Infection, leveraging machine learning methods to turn
voice information into a solid pointer of the infection.
The iterative nature of blunder decrease in SVM
guarantees that the show is both proficient and exact in
classifying the input data. Through this organized
strategy, the framework points to offer a strong apparatus
within the early discovery of Parkinson's Infection,
which is fundamental for opportune therapeutic
mediation.

Module Description

Fig 1:Architecture Diagram of PDD

The chart speaks to the method of detecting

Parkinson's Illness (PD) employing a machine
learning approach with Bolster Vector Machine
(SVM). The flowchart starts with collecting
Parkinson's Malady information, which incorporates
important highlights
Fig 2:Sequence steps for implementation for PD
The graph gives a point by point representation of
the workflow employed for Parkinson's Illness( PD)
discovery exercising machine literacy strategies,
Implementation
centering on converse information examination. The usage of this venture includes a few key steps,
The system begins with the collection of PD beginning with information collection, taken after by
converse Information, which shapes the essential information preprocessing, show preparing, and
dataset. The information at that point gests assessment. The primary step is to assemble a
InformationPre- processing, where any clamor or
significant dataset, such as the UCI Parkinson's Dataset,
gratuitous highlights are expelled, and the converse
information is ready for disquisition, counting which contains voice estimations of people with and
normalization or scaling. This step guarantees the without Parkinson's Infection. These highlights
information is in a applicable arrange for preparing incorporate principal vocal recurrence, jitter, gleam, and
the machine literacy show. other estimations that are successful pointers of the
infection.
Taking after pre-processing, the dataset is separated
into Prepare Information and Test Information. Be Once the information is gotten, it is preprocessed to
that as it may, to guarantee vigorous demonstrate evacuate any non-essential columns (such as persistent
assessment,K-Fold Cross blessing is employed. This names) and to partitioned the highlights (input
strategy corridor the preparing information into' K' information) from the target variable (infection status).
corridor, guaranteeing that the demonstrate is set and The dataset is at that point part into preparing and
approved over distinctive subsets of the information, testing sets, guaranteeing that the demonstrate has
lessening the hazard of overfitting. The test particular information for learning and assessment. To
information is kept away to survey the ultimate make strides the model's execution, include
model's prosecution. standardization is connected, where the information is
normalized so that all highlights contribute similarly to
Amid demonstrate preparing, the hyperactive- the model's expectations.
parameters( similar as learning rate, regularization
quality, etc.) play a significant part in determining the
gests of the machine literacy show. To discover the
most excellent hyperactive- parameters, Network
Look is connected, which efficiently tests distinctive
parameter combinations to optimize the model's
prosecution. Once the parameters are tuned, the Show
Preparing stage commences, where the computation
learns to distinguish designs within the PD converse
information.

After training, the demonstrate is tried on invisible

information from the test set within the Testing stage.
This step is significant for deciding how well the
show generalizes to ultramodern data. Once the
testing is total, the show gests Assessment, where
different prosecution measures (similar as fineness, Fig 3:work flow model
fineness, review, and F1 score) are calculated to
estimate the model’s viability in relating Parkinson's
Infection.
A Bolster Vector Machine (SVM) classifier is chosen
This organized approach guarantees that the show is for preparing due to its viability in taking care of
both reliable and optimized, giving exact vaticinations classification issues. The SVM show is prepared utilizing
with respect to the nearness of Parkinson's Infection the preparing information, and expectations are made
grounded on converse characteristics. on both the preparing and test sets. The precision of the
demonstrate is assessed based on how well it accurately
classifies people as either having Parkinson's Malady or
being solid.
To guarantee a exhaustive assessment, execution
measurements such as precision scores and confusion
matrices are utilized, giving bits of knowledge into
how well the demonstrate performs on concealed
information. Highlight relationship is visualized
employing a heat map, uncovering which highlights
most unequivocally connect with Parkinson's
Malady.

Fig 5:Distribution of parkinson’s disease status

CONCLUSION
In conclusion, this venture effectively illustrates the
Fig 4: Parkinson’s Disease prediction potential of machine learning in identifying Parkinson's
Illness utilizing discourse information. By extricating
In expansion to the model's execution, a prescient and analyzing particular acoustic highlights such as
framework is built, where modern voice information jitter, shine, and essential recurrence varieties, the
can be input to decide whether an person has proposed framework offers a non-invasive and
Parkinson's Malady. This framework permits for real- proficient strategy for early determination. Machine
time expectations based on voice estimations, making learning models like Bolster Vector Machines (SVM)
it a commonsense device for early diagnosis. were utilized to construct a classifier able of recognizing
between people with and without Parkinson's based on
their voice recordings.The comes about show that
analyzing voice information utilizing machine learning
The ultimate step within the usage is sparing the can serve as a profitable apparatus for early discovery of
prepared demonstrate utilizing pickle, empowering Parkinson's Illness, giving an open and cost-effective
the demonstrate to be reused without retraining in arrangement. The system's exactness and execution
future applications. The demonstrate is put away and make it a promising expansion to telehealth stages,
can be stacked afterward to create expectations on empowering farther conclusion and nonstop observing
modern information. This execution system leverages of patients. Whereas the current show appears
Python libraries such as pandas, scikit-learn, and noteworthy potential, future advancements, counting the
librosa, guaranteeing that the approach is both utilize of more progressed calculations, bigger datasets,
effective and adaptable for future upgrades or real- and integration with multi-modal information, may
world arrangement. encourage upgrade its exactness and appropriateness in
real-world healthcare settings.
When we analyze the dissemination of status within
the dataset, we regularly watch:Check of Patients
with Parkinson's Malady (status = 1):This tells us
how numerous people within the dataset have been
analyzed with Parkinson's. Check of Sound People
(status = 0):This appears the number of people
without Parkinson's Illness.
 Future Work [7] J. Mei, C. Desrosiers, and J. Frasnelli, “Machine
Learning for the Diagnosis of Parkinson’s Disease: A
Future enhancements could involve the use of more Review of Literature,” Frontiers in Aging Neuroscience,
sophisticated deep learning methods, such as vol. 13. Frontiers Media S.A., May 06, 2021.
recurrent neural networks (RNNs) or convolutional
[8] M. Shahbakhi, D. T. Far, and E. Tahami, “Speech
neural networks (CNNs). These approaches excel at Analysis for Diagnosis of Parkinson’s Disease Using
recognizing temporal relationships and complex Genetic Algorithm and Support Vector Machine”.
patterns in sequential data, such as speech. Utilizing
these models could lead to a notable increase in the
accuracy of Parkinson’s disease [9] Meireles, J. and Massano, J., 2012. Cognitive
detection.Moreover, integrating multimodal data by impairment and dementia in Parkinson’s disease: clinical
merging speech characteristics with handwriting features, diagnosis, and management. Frontiers in
analysis, gait information, or other clinical data neurology.
could establish a more thorough diagnostic
framework. This would offer a more complete [10] Christoffer Malmberg, "Real-time Audio
understanding of the patient’s condition and Classification on an Edge Device: Using YAMNet and
enhance the model’s diagnostic precision. TensorFlow Lite", 2021.

[11] Himanish Shekhar Das et al., "Early detection of

References Parkinson’s disease using fusion of discrete wavelet
transformation and histograms of oriented
gradients", Mathematics, vol. 10, no. 22, pp. 4218, 2022.

[1] J. S. Almeida et al., "Detecting Parkinsons disease

with sustained phonation and speech signals using [12] A-Hyeon Jo and Keun-Chang Kwak, "Speech
machine learning techniques", Pattern Recognit. Lett., Emotion Recognition Based on a Two-Stream Deep
Learning Model Using Korean Audio
vol. 125, pp. 55-62, 2019.
Information", Applied Sciences, vol. 13, no. 4, pp. 2167,
2023.
[2] C. Quan, K. Ren and Z. Luo, "A Deep Learning
Based Method for Parkinson's Disease Detection Using [13] G. E. Thomas, N. Hannaway, A. Zarkali, K. Shmueli
Dynamic Features of Speech", IEEE Access, vol. 9, pp. and R. S. Weil, "Longitudinal Associations of Magnetic
10239-10252, 2021. Susceptibility with Clinical Severity in Parkinson's
Disease", Movement Disorders, 2024.
[3] W. Wang, J. Lee, F. Harrou and Y. Sun, "Early
Detection of Parkinson's Disease Using Deep Learning
[14] E. Philippou, M. A. Hirsch, P.C. Heyn, E. E. van
and Machine Learning", IEEE Access, vol. 8, pp.
Wegen and H. Darwish, "Vitamin D and Brain Health in
147635-147646, 2020.
Alzheimer and Parkinson Disease", Archives of Physical
[4] Aditi Govindu and Sushila Palwe, "Early Medicine and Rehabilitation, 2024.
detection of Parkin- son's disease using
machine learning", Procedia Computer Science, vol. [15] N. Kumar, R. K. Kaushal and S. N. Panda, "IoT
218, pp. 249-261, 2023, ISSN 1877-0509. Based Smart and Portable System for Remote Patient
Monitoring and Drug Delivery", J. Phys. Conf. Ser., vol.
1950, no. 1, pp. 012017, Aug. 2021.
[5] N. D. Pah, V. Indrawati and D. K. Kumar, "Voice-
based SVM Model Reliability for Identifying
Parkinson's Disease", IEEE Access.

[6] I.Rustempasic and M. Can, “Diagnosis of

Parkinson’s Disease using Fuzzy C-Means Clustering
and Pattern Recognition,” Southeast Europe Journal of
Soft Computing.