Machine Learning Analysis of the Telemonitoring Voice Dataset for Enhanced Parkinson's Disease Severity Prediction

Parkinson's Disease (PD) evolves as a neurodegenerative disorder, advancing progressively and complicating early-stage diagnosis due to its nuanced and complex manifestations. This study introduces an advanced machine learning model to enhance the accuracy of PD severity prediction by analyzing vocal feature variations, a common symptom among PD patients. Utilizing a soft voting ensemble method that integrates several base models—Decision Tree (DT), Naïve Bayes (NB), Logistic Regression (LR), K-NN, and SVM—this research aims to surpass traditional diagnostic methods' performance. Through rigorous preprocessing, including label encoding, feature selection, normalization, and dimensionality reduction, the dataset was optimized for the ensemble model. The proposed model demonstrated superior predictive performance, achieving remarkable accuracy, precision, recall, and F1-scores of 0.8916, 0.8919, 0.8916, and 0.8914 for Total UPDRS scores, and 0.9127, 0.9130, 0.9127, and 0.9127 for Motor UPDRS scores, respectively. These findings significantly outperform conventional diagnostic approaches and highlight the potential of machine learning in revolutionizing PD diagnostics. This study underscores the efficacy of ensemble learning in medical diagnostics and sets the stage for future research to explore its application in clinical settings, aiming to improve early diagnosis and patient outcomes for those afflicted by PD.