Voice-based Parkinson's disease prediction using machine learning

The present work proposes the methods of voice analysis and machine learning in early detection of Parkinson&s;s Disease (PD). Speech recordings of PD patients and healthy individuals were subjected to processing for feature extraction, which included MFCC, jitter, shimmer, and HNR. Various deep learning models were tested together with Random Forests and SVM. The proposed SVM model identified early-stage Parkinson&s;s disease with an accuracy of 92.5%, precision of 91%, and recall of 93%. It once again proved the prospects for non-invasive voice-based detection. This approach is particularly suitable for screening via telemedicine and is scalable and affordable. In this work, explainable AI techniques were incorporated to facilitate clinical trust along with feature selection for efficiency. Future research will be directed toward the expansion of datasets and improvement in generalization to provide early diagnosis and personalized treatment.