Performance evaluation of combined feature selection and classification methods in diagnosing parkinson disease based on voice feature

Parkinson is a disease attacking the nervous system and worsens the work of nervous system over time. This disease is incurable, the therapy existing today is only able to help to relieve the symptoms. Hence, an early diagnose is deemed essential to determine an accurate type of therapy. Parkinson disease can be diagnosed by examining the symptoms apparent to the patient. One of the symptoms is the existence of dysphonia (weakness in voice production) to the patients with Parkinson. This research purposely is to examine the diagnosis of Parkinson disease through the measurement of voice data obtained from UCI repository. The dataset of the voice was initially normalized before conducting the feature selection by means of a number of methods including Correlation-based Feature Selection (CFS), Principal Component Analysis (PCA), Wrapper and conducting without feature selection. The data that has been selected later was classified using four classifiers including Support Vector Machine (SVM), k-Nearest Neighbour (kNN), Bayesian Network and Multi Layer Perceptron (MLP). All of the processes conducted using WEKA 3.6. The result revealed that the highest accuracy was obtained at 98.97% with the sensitivity of 99.32% and specificity of 97.92% from the use of feature selection of Wrapper using kNN classifiers.