Nonlinear Combination of Kernels Using Genetic Algorithm for Improvement of Support Vector Machine Classification Error

Support Vector Machine (SVM), is a powerful machine learning technique widely used for regression and classification. As a classifier, we can use SVM as a linear classifier or kernel based classifier. In case of kernel based classification, the type of kernel function and its parameters affect significantly on classification accuracy. In this paper, we propose a method based on genetic algorithm to obtain a suitable kernel function based on nonlinear combination of conventional kernel functions. We use classification error as our genetic algorithm fitness function which should be minimized. We evaluate the proposed approach using UCI dataset. Results show that this nonlinear combination can improve SVM true classification rate