Performance comparison of heterogeneous classifiers for detection of Parkinson's disease using voice disorder (dysphonia)

Speech signal processing and its recognition system have gained a lot of attention from last few years due to its widespread application. In this study, we have conducted a comparative analysis for effective detection of Parkinson's disease using various machine learning classifiers from voice disorder known as dysphonia. To investigate robust detection process, three independent classifier topologies were applied to distinguish between PD patient and healthy individual, and to make a comparison of the results. The classifiers used here are Random Tree (RT), Support Vector Machine (SVM) and Feedforward Back-propagation based Artificial Neural Network (FBANN). To validate the overall classification with acceptable error rate, a 100 times repeated 10-fold cross validation analysis has been carried out for all classifiers. With optimized statistical parameters and using selective feature set, the proposed scheme has achieved up to 97.37% recognition accuracy. FBANN classifier outperformed than the others. Considering the classification accuracy, sensitivity, specificity and the area under the receiver operating characteristic (ROC) curve, all classifiers achieved better than chance level. The proposed modality and computational process may clinically effective, viable, noninvasive, powerful technique to develop decision support system (DSS) for remote diagnosis of neurodegenerative disorders at early stage with propitious results.