A random forest-based methodology for superior performance in voice-assisted early detection and remote monitoring of Parkinson’s disease

Abstract

Parkinson’s disease (PD) is a progressive neurological disorder characterized by motor symptoms such as shivering, rigidity, and slow movement. While no cure is currently available, drug treatments can significantly manage these symptoms, which tend to worsen as the disease advances. Of particular research interest is the early sight of PD, as speech impairment is a notable early sign that often appears years before typical motor symptoms, such as the initial slight trembling or reduced arm swing when walking. Analyzing speech and voice patterns thus presents a promising, non-invasive avenue for both early prognosis and remote, ongoing evaluation of the clinical status of PD, presenting an effective way to monitor patients in their domestic context. This study aims to develop and evaluate a machine learning-based approach to enhance the precision of predicting PD in individuals presenting with speech difficulties. The investigation employed a comprehensive set of machine learning classifiers, including Support Vector Machines (SVM), K-Nearest Neighbours (KNN), Random Forests (RF), Decision Trees (DT), and Ada-Boost, to identify indicative patterns within speech and voice data. The empirical probe demonstrated that the RF classifier significantly outperformed the other techniques, achieving an accuracy of 96.61%, a precision of 0.96, and a recall of 1.0. These results represent a substantial enrichment in both attainment and intricacy related to prior studies on voice-based PD identification using the same dataset. This work underscores the potential of the proposed machine learning methodology, specifically using the RF classifier, to effectively avail in the early sight and remote monitoring of Parkinson’s disease.