ABSTRACT : A neurodegenerative disease that affects the brain’s neurological, physiological, and behavioral systems, Parkinson’s disease (PD) is difficult to diagnose early because of its subtle symptoms. Slowness of movement, or bradykinesia, is a hallmark PD symptom that usually first appears in middle adulthood and gradually worsens. Verbal communication impairment is one of the major effects of Parkinson’s disease. Support Vector Machine (SVM), Naïve Bayes, k-Nearest Neighbors (k-NN), and Artificial Neural Networks (ANN) were among the supervised classification techniques used in this study to identify speech-related abnormalities linked to Parkinson’s disease. Using the combined predictive strength of several models that were trained separately, ensemble learning techniques were used to improve diagnostic robustness. In particular, the following four ensemble algorithms were compared: majority voting, weighted voting, bagging, and AdaBoost. The 195 speech signal samples in the dataset included 48 samples from healthy controls and 147 samples from people with Parkinson’s disease. The results of the experiments showed that ensemble approaches performed noticeably better than individual classifiers. Using a majority voting ensemble that integrated Decision Tree, k-NN, and SVM classifiers, the highest accuracy of 95.3 percent was attained, highlighting the effectiveness of ensemble techniques in improving PD detection of speech patterns.

Keywords – neurodegenerative disorder, Parkinson’s disease; machine learning, disease prediction, adaboost, bagging, majority voting, soft voting, ensembled.