Preliminary Study on Machine Learning Application for Parkinson's Disease Diagnosis

Abstract

Early detection for Parkinson's Disease (PD) can be realized by investigating the speech abnormalities of the patient. Utilizing machine learning approach, PD can be well diagnosed by investigating its speech features. Oxford Parkinson's Disease (OPD) dataset, containing pieces of PD patients' speech and normal speech was used in this study. The investigated algorithms that were tested are Support Vector Machine, K-Nearest Neighbor, Linear Discriminant Analysis, Gradient Boost, Multi-layer Perceptron, and Decision Tree. The performance evaluation of all these methods is based on accuracy, precision, recall, and F1 score. Based on the evaluation, the most suitable algorithm for PD case is Multilayer Perceptron with the accuracy of 95.92% without data scaling.