Enhancing Parkinson’s Diagnosis: Unleashing the Potential of Machine Learning via Vocal Features Fusion

Parkinson’s disease (PD) is a degenerative brain disease that worsens with age, causing areas of the brain to weaken. The main focus of this study has been on the characteristics of the voice modality with a wide range of features including vocal fold, MFCC (Mel-Frequency Cepstral Coefficients), baseline features, wavelet features, temporal frequency features and Tunable Q-Factor Wavelet Features. The proposed models introduced dimensionality reduction techniques such as PCA (Principal Component Analysis) and ICA (Independent Component Analysis) to improve the quality of the data with different Machine Learning (ML) classifiers (K-Nearest Neighbors, Support Vector Machines, Naive Bayes etc.) and their ensemble technique for classification of PD. Moreover, an ensemble approach comprising multiple Deep Neural Networks (DNNs) models was employed to assess the efficacy of the proposed methodology. Through the incorporation of a weighted average, the DNN ensemble exhibited a notable accuracy of $\mathbf{8 8. 7 4 \%}$ in effectively classifying Parkinson’s patients. However, the fusion of MFCC and vocal fold feature sets yielded exceptional results, showcasing a robust accuracy of $\mathbf{9 8. 6 8 \%}$ with the utilization of a traditional ML ensemble model.