Kernal Width Selection for SVM Classification

A rule-based meta-learning approach for automatic radial basis function (RBF) kernel and its parameter selection for Support Vector Machine (SVM) classification and it is observed that this methodology provides significant improvement of computational time as well as accuracy in some specific cases.

Abstract

The most critical component of kernel-based learning algorithms is the choice of an appropriate kernel and its optimal parameters. In this paper, we propose a rule-based meta-learning approach for automatic radial basis function (RBF) kernel and its parameter selection for Support Vector Machine (SVM) classification. First, the best parameter selection is considered on the basis of prior information of the data with the help of Maximum Likelihood (ML) method and Nelder-Mead (N-M) simplex method. Then, the new rule-based meta-learning approach is constructed and tested on different sizes of 112 datasets with binary class as well as multi-class classification problems. We observe that our rule-based methodology provides significant improvement of computational time as well as accuracy in some specific cases.