EXAMINING ML ALGORITHMS FOR PARKINSON'S DETECTION THROUGH SPEECH DATASETS : A COMPARATIVE ANALYSIS

Abstract

Parkinson's disease or PD, the most well-known neurological condition impacting the human neurological system, causes dopamine-producing neurons in the midbrain to degenerate. It is a primary concern to detect PD in its early stages to slow down its progress by engaging patients in early medical therapies and foster a better quality of life for them. Although new research appears to indicate that majority of the PD patients experience speech impairments in the early stages of the disease, the primary impacts of PD are on motor and cognitive function. Within the framework of this study, a number of machine learning (ML) models, including Principal Component Analysis (PCA), Random Forest (RF), Gaussian Naïve Bayes (GNB), K-Nearest Neighbours (KNN), Decision Tree (DT), Logistic Regression (LR), Extreme Gradient Boosting (XGB), and Support Vector Machine (SVM), have been comparatively analysed on two different speech datasets consisting multiple attributes, using three different approaches for classification. The models were assessed and evaluated, for their efficiency in PD classification, using different scoring metrics such as accuracy, precision, recall, and Fl-score. Here, we discovered that the XGB and SVM models of the second approach—where the data was oversampled—were the most efficient models. XGB demonstrated 98.30% accuracy and 96.67% precision with Dataset 1 while SVM achieved 97.8% accuracy and 99.1% precision with Dataset 2. They also depicted maximum area under the curve for ROC curve, highlighting their capability to discriminate between true positives and true negatives. The highest degree of accuracy and precision in the early detection of PD has been rendered attainable by ML algorithms. When trained on an extensive set of data, these additionally possess the potential to offer 100% accuracy, or clinical-grade accuracy, through hyper-parameter optimisation.