Comparing two SVM models through different metrics based on the confusion matrix

Abstract

Support Vector Machines (SVM) are an efficient alternative for supervised classification. In the soft margin SVM model, two different objectives are optimized and the set of alternative solutions represent a Pareto-front of points, each one of them representing a different classifier. The performance of these classifiers can be evaluated and compared through some performance metrics that follow from the confusion matrix. Moreover, when the SVM includes feature selection, the model becomes hard to solve. In this paper, we present an alternative SVM model with feature selection and the performance of the new classifiers is compared to those of the classical soft margin model through some performance metrics based on the confusion matrix: the area under the ROC curve, Cohen’s Kappa coefficient and the F-Score. Both the classical soft margin SVM model with feature selection and our proposal have been implemented by metaheuristics, given the complexity of the models to solve.