A multiple classifier selection method with self-adaptive preferences

Clustering and Selection (CS) is a common method of multiple classifier selection. But the method judging an input sample belong to a certain area just by the shortest distance has some unilateralism. Therefore, a dual selection method based on clustering is proposed. In the method, multiple clusters are selected for a test sample and the classifier with the best weighted average performance is chosen. The chosen classifier is compared with the best classifier in the nearest cluster and the better one are used to classify the test sample. The main parameter in the method is self-adaptively selected according to the prior information of the training samples. Experiments were done on the datasets of KDD´99 and UCI to compare the proposed method with the CS method, and the experimental results show the presented method has a better classification performance.