A new classification algorithm based on ensemble PSO_SVM and clustering analysis

Abstract

Aiming at the existing problems of support vector machine ensemble, such as strong randomicity, larger scale of training subsets size and high complexity of ensemble classifier, this paper put forward a novel SVM ensemble construction method based on clustering analysis. Firstly, the samples are clustered into several clusters according to their distribution with rival penalty competitive learning algorithm(RPCL). Then a small quantity of representative instances are chosen as training sets and training SVM that adopt self-perturbation in population convergence speed. Finally Ensemble improvement SVM is constructed by relative majority voting. Man-made data are used to test C_PSOSVM. Experiment result illustrate that the algorithm can improve ensemble SVM classification precision, reducing time-space complexity compared with Bagging, Adaboost.