Adaptive Variational Mode Decomposition Based Parkinson Disease Detection with Entropy-Based Features

Detection of Parkinson's disease (PD) presents significant diagnostic challenges, especially in its early stages. This study introduces a new method for the detection of PD from speech-based audio signals by leveraging Adaptive Variational Mode Decomposition (AVMD) and Random Forest (RF) classifier. The method includes advanced entropy-based features, specifically Renyi, spectral, and permutation entropy, to improve the analysis of audio signals for diagnostic purposes. The methodology involves preprocessing vocal audio data from PD patients and healthy controls, followed by feature extraction using AVMD to handle the non-stationary nature of speech. The entropy-based features are then utilized to capture the complex dynamics within these audio signals. Classification is performed using an RF classifier, which is rigorously evaluated through k-fold cross-validation to ensure the robustness and reliability of the diagnostic model. The study's results are promising, with the RF classifier achieving a peak Unweighted Average Recall (UAR) of 78.30% for vowel E when considered independently, and an UAR of 71.88% for the combination of all five vowels. These findings indicate significant potential for this method in clinical settings.