Parkinson's Disease Detection Based on Vocal Biomarkers and Machine Learning Approach

Abstract

Parkinson's disease (PD) must be identified early to provide prompt treatment and better patient outcomes. This study looks into using machine learning methods in conjunction with vocal biomarkers to identify Parkinson's disease early. Vocal data was analyzed using four techniques: Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Random Forest, and Logistic Regression algorithms. The third strategy, which used the Synthetic Minority Over-sampling Technique (SMOTE) for data balance, performed exceptionally well and stood out among the others. Utilizing SMOTE for dataset balancing, the Random Forest algorithm produced an exceptional 98.3 % accuracy. The study highlights the effectiveness of voice biomarkers, examines the various approaches used, and draws comparisons between them.