You are on page 1of 17

Hybrid Model for Detecting

Parkinson’s Disease using


speech data
• Group ID : 17
• Members : Malay Thakkar,
Viraj Lande,
Neelanjaan De,
Bharat Dedhia,
Guide : Prof. Snigdha Bangal
Problem Definition

Detection of Parkinson’s Disease using stacking classifiers


made from Machine Learning Algorithms such as Support
Vector Machine, K-nearest neighbor and XGBoost. this system
will be deployed through a web application. This will also make
use of Synthetic oversampling of the minority class and use of
a feature selection algorithm while retaining the most
important ones that contribute to the accuracy of the model.
Introduction
• Parkinson's disease is a neurodegenerative disorder that
affects movement and is characterized by symptoms such as
tremors, rigidity, and difficulty with balance and
coordination.
• Doctors usually diagnose the disease by taking a person’s
medical history and performing a neurological examination,
These include: DaTscan, Motor function tests, Genetic testing
all together to confirm the result.
• Using speech data from subjects is expected to help the
development of a noninvasive diagnostic.
Literature Survey (papers referenced)
Literature Survey (papers referenced)
Dataset Description
How is Data collected:
• Recording is achieved by a Trust MC-1500 microphone with a
frequency range between 50 Hz and 13 kHz
• The microphone is set to 96 kHz, 30 dB and placed at 10 cm
distant from the subject
• patients are asked to say only the sustained vowels “a” and
“o” three times
• Features are extracted from the voice samples using acoustic
analysis software.
Dataset Description
The publicly available dataset from the UCI repository has been used in this study( 195 instances)
Scope
Functional Requirements (for webapp)

• Data-collection fields: Collection of relevant data.


• Validation: Validation of the user input
• Prediction: Can make predictions on new, unseen data.
• Autofill: corresponding fields are auto-filled to another model while
filling inputs of one model, to aid accessibility.
• Result Display: Results of the diagnosis in a Lucid manner, along with
other relevant information (if any).
• Other Accessibility: mobile compatible, light/dark/wide mode.
Scope
Non - Functional Requirements (Web app)
• Performance: Real-Time Detection, Low Latency, and High Accuracy.
• User-friendliness: Easy to use, User-Friendly interface
• Privacy: The system should not collect user data without their
permission
• Reliability: No Frequent downtimes.
• Maintenance: Simple Process for Maintaining the Web Application.
Technologies to be used
• ML Model: - NumPy, Web App: - Streamlit,
Pandas,
Matplotlib Python
Seaborn,
Scikit-learn

**(may subject to change)


Overview of Implementation
Results

Selected Features K- Accurac f1- precisio roc_au


fold scores Recall y score n c
SVM 93.3 93.22 93.2 93.77 98.14
KNN 92.88 92.83 92.8 93.4 95.84
XGBoost 93.71 93.66 93.6 94.28 98.43
Proposed Method
(Stacking classifier) 94.55 94.5 94.46 95.08 98.36

Full Features K-fold Recal f1- precisio roc_au


scores l Accuracy score n c
SVM 96.66 96.63 96.62 96.97 99.49
KNN 89.47 89.37 89.11 91.59 98.39
XGBoost 93.75 93.64 93.61 94.31 98.28
Proposed Method
(Stacking classifier) 97.08 97.05 97.04 97.3 99.42
Results
Results
Conclusion
This project aims to use speech signals for early detection of
Parkinson's disease, as early identification is crucial for
prolonging a patient's life. While there is no common feature
selection methodology or classifier for medical datasets, a
hybrid system has been developed with a 97.05% accuracy
rate. This system can provide a second opinion for Parkinson's
diagnosis but should not replace medical experts. The
system's effectiveness may vary depending on the chosen
feature selection methodology and classifier.
Future Scope
• Detection of Parkinson’s disease can be done from a
diagnosis system by handwritten drawings as slowness and
tremors is also an early symptom of Parkinson’s which
adversely affects the patient’s handwriting.
• Different tools and techniques may be tried to yield different
results such as different SMOTE variants.
• Alternative methods for extracting voice features may be
explored.
References
1. https://www.kaggle.com/datasets/debasisdotcom/parkinson-disease-detection
2. Lamba, R., Gulati, T., Alharbi, H.F. et al. A hybrid system for Parkinson’s disease diagnosis using
machine learning techniques. Int J Speech Technol 25, 583–593 (2022)
3. Z. Karapinar Senturk, Early Diagnosis of Parkinson’s Disease Using Machine Learning
Algorithms, Medical Hypotheses (2020)
4. Salim Lahmiri, Amir Shmuel, Detection of Parkinson’s disease based on voice patterns ranking
and optimized support vector machine, Biomedical Signal Processing and Control
5. S. A. Mostafa, A. Mustapha, M. A. Mohammed et al., Examining multiple feature evaluation and
classification methods for improving the diagnosis of Parkinson’s disease, Cognitive Systems
Research
6. Ashena Gorgan Mohammadi, Pouya Mehralian, Amir Naseri, Hedieh Sajedi, Parkinson's disease
diagnosis: The effect of autoencoders on extracting features from vocal characteristics
7. K. Polat, "A Hybrid Approach to Parkinson Disease Classification Using Speech Signal: The
Combination of SMOTE and Random Forests," 2019 Scientific Meeting on Electrical-Electronics
& Biomedical Engineering and Computer Science (EBBT)

You might also like