Abstract
Parkinson’s Disease (PD) is a serious chronic illness known to slow the motor function of a human being as it affects movement and speech. There are significant benefits of early diagnosis of the disorder and it is essential that PD is diagnosed as early as possible. This paper assesses the applicability of stateof-art vocal features which are Vocal Tract Length Normalization (VTLN), Empirical Mode Decomposition (EMD), and Continuous Wavelet Transform (CWT) in combination with the recent Machine Learning (ML) algorithm for the identification of PD. Hence, we performed the performance assessment of different types of models such as Explainable Boosting Machine (EBM), Fast and Lightweight AutoML (FLAML), as well as NGBoost using 195 recorded vocal data sets. EBM was found to be the model with the highest accuracy of 86. 67%, and the AUC was 87. 33% for the same model and FLAML demonstrated a sensitivity score of 100%. The results of this work shed light on how sufficient analysis of the vocal material may be effectively combined with the contemporary ML algorithms to enhance the accuracy of PD identification.