Automatic Recognition of Descriptors helping to Cause Diabetes in Algeria

Abstract

The computer aided medical diagnosis systems can use a great number of very important medical data in order to help doctors in detecting different pathologies. We assume that the grater data we have, the more we facilitate and ameliorate the quality of classification. However, the classification quality does not directly depend on the size of the available database but it rather depends on its pertinence. For this, the purpose of this paper is to two different problems. The first one is the selection of the pertinent descriptors that help causing diabetes using a Random Forest feature selection approach. The second is the combination of several different machines learning algorithms (Support Vector Machine (SVM), K-Nearest Neighbor (KNN), the Multilayer Perceptron (MLP) and two Decision tree based classifiers (Classification And Regression Tree (CART), and Random Forests) in order to classify type 2 diabetic patients. We used also a majority voting method between the proposed five classifiers. In our paper, we selected an experimental database composed of 625 patients, each of whom being represented by 31 descriptors. These patients were selected in various private clinics and hospitals in western Algeria.