Machine learning approach for classification of Parkinson disease using acoustic features

Abstract

Parkinson’s disease (PD) is common disorder for many people and is not easy to diagnose. It is a neurological disorder. The authors proposed a novel approach using data partitioning with feature selection algorithm Principal component analysis (PCA) for Parkinson’s disease classification. In the proposed approach, the dataset has been divided into three equal parts and validated two-class (healthy and Parkinson’s disease) for individual data with different classifiers based on acoustic features. To improve performance of classifying algorithms Principal Component Analysis (PCA) has been used. The minority and majority classes were obtained by applying the data set partition approach to the dataset of healthy and Parkinson’s disease subjects. The three equal partitions of were composed for healthy (first case), and then for PD class (second case). PCA was used for features selection. We used three different classifiers to classify all data partitions, including the weighted k-NN (nearest neighbour, wkNN), Logistic Regression (LR), and Medium Gaussian Kernel support vector machine (MGSVM). The classification accuracy of 74.2%, 85.0% and 82.1% achieved using Logistic algorithm, SVM with Gaussian, and weighted k-NN classifiers. The combination of classifiers, data partition and feature selection (first case) achieved classification accuracy of 80%, 87.63% and 89.23% respectively. In the second case, 85.2%, 89.36% and 90.3% accuracy with data partition and feature selection are obtained respectively. The results show that the proposed methodology could be used for Parkinson’s disease classification.