An ensemble SVM using entropy-based attribute selection

In order to improve the generalization performance of support vector machine (SVM), a kind of ensemble SVM using an entropy-based attribute selection method was proposed. An entropy metric based on similarity between objects was designed to evaluate the importance degree of each attribute and so as to obtain a set of important attributes. Based on the set of important attributes, the Bagging method was used to generate sub-SVMs and then the majority voting rule was adopted to obtain the final ensemble result of all sub-SVMs. The proposed ensemble method can avoid destructing the attribute relativity or attribute dependence by selecting an attribute subset from original attribute space randomly. The performance of single SVM can be improved and the diversity between sub-SVMs can also be guaranteed. Simulation results on UCI testing datasets show that the proposed ensemble method can improve the classification precision of SVM and make the ensemble SVM has better generalization property.