Redesigning a NSGA-II metaheuristic for the bi-objective Support Vector Machine with feature selection

Abstract

The Support Vector Machine is a well-known technique used in supervised classification. Feature selection offers several benefits but also adds complexity to the problem. In this paper, we consider the soft margin SVM and given that two different objectives are considered simultaneously, obtaining the Pareto front , or at least a good approximation of it, gives the decision-maker a wide variety of solutions and has several advantages over having only one solution. The only metaheuristic that has been developed to give an approximation of such a front is a NSGA-II based technique. However, the design of such technique presents some limitations that are analyzed in this paper. We present a new metaheuristic that has been completely redesigned in order to overcome those drawbacks. We compare both techniques through an extensive computational experiment that demonstrates the superior efficiency of the new technique.