Radial Basis Function Kernel Parameter Optimization Algorithm in Support Vector Machine Based on Segmented Dichotomy

Abstract

By analyzing the influences of kernel parameter and penalty factor for generalization performance on Support Vector Machine (SVM), a novel parameter optimization algorithm based on segmented dichotomy is proposed for Radial Basis Function (RBF) kernel. Combine with Segmented Dichotomy(SD) and Gird Searching(GS) method, a composite parameter selection, SD-GS algorithm, is structured for rapid optimization of kernel parameter and penalty factor. UCI Machine Learning database is used to test our proposed method. Experimental results have shown that performance on parameter selection is better than traversal exponential grid searching. Thus, the optimized parameter combination of SD-GS algorithm enables RBF kernel in SVM to have higher generalization performance.