Diagnosis of Parkinson's disease using genetic algorithm and support vector machine with acoustic characteristics

Abstract

Parkinsons disease (PD) is a neurological illness which is usually accompanied by dysphonia. In this paper, we proposed a diagnosis method of PD using genetic algorithm (GA) and support vector machine (SVM) based on the acoustic characteristics of Parkinson's patients for improving the diagnosis accuracy. Firstly, A comparison study of classifiers' performance was conducted between SVM and decision tree (C4.5), K nearest neighbor (KNN), and probabilistic neural network (PNN). The results showed SVM outperformed the three classifiers. Secondly, the normalization of feature vector was adopted before training SVM. The prediction accuracy of SVM was improved from 91.8% to 96.4%. Thirdly, GA was applied into feature selection for improving the performance of SVM. The result showed the accuracy of SVM further increased to 99.0% and the dimension of feature vector decreased from 22 to 10. The study demonstrated that the combination of GA and SVM is a practical method of diagnosis PD.