Based on Grid-Search and PSO Parameter Optimization for Support Vector Machine

When using SVM to solve practical problems, the selection of the kernel function and its parameters plays a vital role on the results of good or bad, and only need to select the appropriate kernel function and parameters to get a SVM classifier with good generalization ability. RBF kernel function gets the most widely used, and there are only two parameters, which are the C and gamma. This paper discusses the parameter selection method of PSO and grid-search respectively. The grid-search method need to search for a long time, while PSO is easy to fall into local solution, for these shortcomings, an improved method combining PSO and the grid-search method is proposed in this paper. The comparative experiment on ORL results show that the proposed method has faster recognition speed and higher recognition accuracy than the grid-search method. This method has higher recognition accuracy than the method with the PSO alone, and it can effectively avoid the algorithm into a local solution.