Abstract: Currently, Multiple Classifier System (MCS) attracts more and more attentions and has become one of the research hotspots in the pattern recognition field. Classifier selection is a commonly used strategy for MCS to achieve the final decision. A classifier selection method based on clustering and weighted mean is proposed in this paper. In the method, multiple clusters are selected according to the distances between cluster centers and the input sample. Then, the average performance of each classifier on selected clusters is calculated. The best classifier on the nearest cluster and the classifier with the best average performance are picked out. According to the reliability of their outputs which are estimated by confusion matrix, one of them is selected to make the final decision of the system. A number of benchmark data sets from KDD’99, UCI and ELENA database were used to evaluate the proposed method. It can be seen form the experimental results that the proposed method performs well.
Keywords: Multiple classifier system, classifier selection, clustering, weighted mean
