A Comparative Study of Machine Learning and Deep Learning Models for Early Detection of Parkinson's Disease Using Voice Features

Abstract

The early detection of Parkinson’s Disease (PD) is a critical challenge, especially since vocal changes often emerge as an early, non-invasive symptom. This study aims to evaluate and compare the performance of Machine Learning (ML) and Deep Learning (DL) algorithms in classifying PD patients from healthy individuals, relying on a standardized set of quantifiable acoustic features (such as Jitter, Shimmer, HNR, and PPE). The CRISP-DM framework was adopted to ensure a robust and reliable methodology. Three distinct classification models were selected for comparison: Support Vector Machines (SVM) with an RBF kernel, Random Forest (RF), and a Deep Neural Network (DNN). The models were trained and evaluated using rigorous performance metrics pertinent to the medical context, including Accuracy, Recall, Precision, and F1-Score. The results, which will be discussed, demonstrate the identification of the most effective model in achieving a high balance between sensitivity and specificity, providing clear insights for developing non-invasive, AI-based diagnostic systems to aid in the early detection of Parkinson's Disease.