A Fast Parameters Selection Method of Support Vector Machine Based on Coarse Grid Search and Pattern Search

Abstract

Parameters selection of support vector machine (SVM) is a key problem in the application of SVM, which has influence on generalization performance of SVM. The commonly used method, grid search (GS), is time-consuming especially for very large dataset. By using coarse grid search and pattern search (PS) to select kernel parameters and penalty factor, a fast method of parameters selection of SVM based on hybrid optimization strategy is proposed in this paper. The proposed method adequately combines the advantages of GS and PS. The experiment results demonstrate that this proposed method can not only improve accuracy and generalization performance of SVM, but also save much more time.