Sound Classification of Parkinsonism for Telediagnosis

Abstract

In recent years the usage of speech-based data for classification of Parkinson disease (PD) has commonly been assumed as non-invasive and effective mode of classification. As a result, an increased interest is observed in speech pattern analysis techniques appropriate for Parkinsonism with the aim of developing predictive tele-monitoring and telediagnosis models. In this research work a method for classification of PD patients is proposed by using different ensemble methods. For this purpose, a set of selected acoustic features, related to frequency, pulse, voice, pitch and harmonicity parameters, are extracted from PD patients' speech dataset and different classifiers (individual, ensemble and combination of ensemble methods) are applied on these 15 extracted features and achieved the overall accuracy of 97.6%. Research aims to early and accurate detection ofdisease in PD patients. Classification accuracies, sensitivity and specificity achieved from the proposedexperimental setup of these ensemble methods are higher than the existing methods using individualclassifiers.