Publication

Abstract : Parkinson's Disease (PD) is a neurodegenerative disease that severely affects the elderly population in multiple dimensions. Early detection of PD employing the conventional techniques is challenging. In this work, we investigate the impact of various entropy measures of Intrinsic Mode Functions (IMFs), derived from speech using Empirical Mode Decomposition (EMD), for the identification of Parkinson's disease. We evaluated the effectiveness of the proposed features using the Italian Parkinson's dataset. We decomposed each preprocessed audio sample into IMF using EMD and extracted four types of entropy features: Shannon, Permutation, Approximate, and Fuzzy entropy of the IMFs. These features capture the irregularities and complexity present in the voice of a person with PD. We evaluated the performance of these features using Machine Learning (ML) models such as Random Forest, KNN, Logistic Regression, and SVM. To ensure the reliability, we tested all models using 5-fold cross-validation. Among the ML models we tried, the Random Forest classifier with Shannon entropy outperforms others in PD detection.