Improved SVM-RFE feature selection method for multi-SVM classifier

Efficient feature selection is a key point in pattern classification. In this paper, we propose an improved feature selection method utilizing support vector machine approach based on recursive feature elimination (SVM-RFE) for multi SVM classifier. This method uses class interval in SVM algorithm as the evaluation criterion, and eliminate features in a recursive way. And in this procedure, obtaining the optimal SVM is a foundation for feature selection. To solve this problem, chaos particle swarm optimization (CPSO) algorithm is applied. At last, the proposed method is employed in classification experiments based on UCI repository, and the approving results show the availability of it.