Parkinson's disease classification using machine learning algorithms: performance analysis and comparison

Abstract

Detection of Parkinson's disease remains challenge for physicians, especially, in the clinical field due to the difficulty of cure. Thus, algorithms of classification have the main role in the assessment of this neurodegenerative disorder. In this paper, we focus on the analysis and the evaluation of nine Machine Learning Algorithms (MLA), namely Support Vector Machine (SVM), Logistic Regression, Discriminant Analysis, K-Nearest Neighbors (KNN), Decision tree, Random Forest, Bagging tree, Naïve Bayes, and AdaBoost. Classification algorithms were applied to a Parkinson's dataset of 240 speech measurements with 44 features using several evaluation parameters to establish the efficiency score of each classifier. We found that the KNN classifier yielded the highest accuracy rate of 97.22% and F1-score of 97.30%.