This project investigates the voice data in patients with Parkinson’s disease, obtained from an open database (UCI Machine Learning Repository). The dataset contains 22 voice features in 195 individuals. Some key vocal features include subtle pitch and amplitude instability, noise-to-harmonics ratio and complex vocal dynamics. The project utilizes RBF SVM to classify individuals into Parkinson’s and healthy groups (Fig 13). The cross-validation result shows 96.6% accuracy of performance. Post-hoc SHAP analysis identifies the most contributing voice features: 1) pitch variation; 2) pitch period entropy; 3) average fundamental frequency: 4) difference between the highest amplitude frequency and the fundamental frequency; and 5) detrended fluctuation.