International Journal of Science, Engineering and TechnologyAuthors: Ishak Gauri, Himanshu Shrivastava, Irah Khan, Hritik Raj, Sohan Lal
Abstract: Parkinson's disease (PD) can be described as a debilitating disorder in which there is disruption to the dopaminergic pathway and associated with various motor dysfunctions. More importantly, one of the key but poorly exploited aspects of PD diagnosis is that vocal dysfunction occurs years before any motor symptoms. In this paper, a novel computational system for the diagnosis of PD through voice analysis is proposed. The proposed approach consists of feature extraction and the application of classification methods such as SVM, RF, KNN, and XG-Boost. Acoustic features including sustained phonation's jitter, shimmer, HNR, and Mel-Frequency Cepstral Coefficients (MFCC) are extracted and fed into machine learning algorithms. Linear kernel SVM provided the best result among all classifiers with an accuracy of 94.87% for training and 87.18% for testing with 195 data instances. Moreover, a web application for real-time PD diagnosis was developed with Flask backend and React frontend. It was shown that the biomarkers of voice signals are promising ways to non-invasively diagnose PD without much cost.
DOI: http://doi.org/10.5281/zenodo.20397202
