A general frame for building optimal multiple SVM kernels

Abstract

The aim of this paper is to define a general frame for building optimal multiple SVM kernels. Our scheme follows 5 steps: formal representation of the multiple kernels, structural representation, choice of genetic algorithm, SVM algorithm, and model evaluation. The computation of the optimal parameter values of SVM kernels is performed using an evolutionary method based on the SVM algorithm for evaluation of the quality of chromosomes. After the multiple kernel is found by the genetic algorithm we apply cross validation method for estimating the performance of our predictive model. We implemented and compared many hybrid methods derived from this scheme. Improved co-mutation operators are used and a comparative study about their effect on the predictive model performances is made. We tested our multiple kernels for classification tasks but they can be also used for other types of tasks. © 2012 Springer-Verlag.