Maximum robustness criterion on kernel selection of support vector machine

Kernel selection is a key issue of support vector machine (SVM), however, for the existing criterions of kernel selection in the literature, there are some limitations in generalization ability , computational cost or fair comparison. This paper proposes a novel criterion of kernel selection for SVM named maximum robustness criterion. The properties of maximum robustness criterion guarantee that it has an excellent generalization ability. Moreover, the maximum robustness criterion is fair for different kernel functions because it is calculated in input space and has low computational cost. The experiment results on some benchmark data sets show that it not only achieves almost as good generalization ability as minimum misclassifying rate of k-fold cross validation, but also has less computational cost.