A comparative study of K-Nearest Neighbour, Support Vector Machine and Multi-Layer Perceptron for Thalassemia screening

Abstract

In this paper, we investigate the feasibility of two typical techniques of Pattern Recognition in the classification for Thalassemia screening. They are the Support Vector Machine (SVM) and the K-Nearest Neighbour (KNN). We compare SVM and KNN with a Multi-Layer Perceptron (MLP) classifier. We propose a two-classifier system based on SVM. The first layer is used to differentiate between pathological and non-pathological cases while the second layer is used to discriminate between two different pathologies (α-thalassemia carrier against β-thalassemia carrier) from the first output layer (pathological cases). Using the parameters sensitivity (percentage of pathologic cases correctly classified) and specificity (percentage of non-pathologic cases correctly classified), the results obtained with this analysis show that the MLP classifier gives slightly better results than SVM although the amount of data available is limited. Both techniques enable thalassemia carriers to be discriminated from healthy subjects with 95% specificity, although the sensitivity of MLP is 92% while that of SVM is 83%.