sélection de variables pour la reconnaissance de la maladie de parkinson

Abstract

The selection of variables in classification arises generally, when the number of variables is high. In this study, we propose some innovative methods to reduce the initial dimension of data in order to select the whole pertinent variables for a supervised classification. Our research work fits into domain help of medical diagnosis. Therefore, in this manuscript, we are interested in the detection and recognition of Parkinson’ disease. Our first contribution is concerned with proposing two supervised classifiers: the SVM and the KNN in order to evaluate subset’s pertinence. Our second contribution consists in proposing two distinct classification approaches by means of two variable selection methods, namely: ‘relieff’ and ‘fisher’ that are intended for the selection of the most pertinent variables. Our experiments have led us to the identification of Parkinson’s disease when using two supervised classifiers based upon variables’ selection.