MACHINE LEARNING-BASED EARLY DETECTION OF PARKINSON’S DISEASE USING VOICE ANALYSIS

Abstract

This study investigates the potential of machine learning techniques, specifically k-Nearest Neighbours (KNN)and Support Vector Machines (SVM), for detecting early-stage cases of Parkinson’s disease (PD) using voice data. Leveraging a dataset from the UCI Machine Learning Repository, which consists of 147 phonetic samples from PD patients and 48 from healthy controls, the methodology involved data preprocessing, feature selection using a genetic algorithm, and handling class imbalance with the Synthetic Minority Oversampling Technique (SMOTE). Dimensionality reduction was performed using Principal Component Analysis (PCA), retaining the most informative features. Both classifiers were trained and validated using stratified 10-fold cross-validation to ensure robust performance evaluation. The KNN classifier achieved an accuracy of 96.11%, with high precision (95.87%), recall (94.76%), and an AUC-ROC of 0.97, indicating superior discriminatory power. The SVM classifier also demonstrated strong performance, with an accuracy of 94.57%, precision of 93.68%, recall of 92.35%, and an AUC-ROC of 0.95. The results shows that KNN model is effective in distinguishing PD patients from healthy individuals using non-invasive phonetic data. The study underscores the phonetic analysis as a reliable biomarker for early PD detection, offering a promising alternative to current diagnostic methods that rely heavily on clinical observation. Future work should focus on validating these findings with larger, more diverse datasets and integrating additional data types to further improve diagnostic accuracy and clinical applicability. The findings support the development of accessible and early diagnostic tools that could significantly enhance patient’s quality of life.