A rough margin-based one class support vector machine

Abstract

We propose a rough margin-based one class support vector machine (Rough one class SVM) by introducing the rough set theory into the one class SVM, to deal with the over-fitting problem. We first construct rough lower margin, rough upper margin, and rough boundary and then maximize the rough margin rather than the margin in the one class SVM. Thus, more points are adaptively considered in constructing the separating hyper-plane than those used in the conventional one class SVM. Moreover, different points staying at the different positions are proposed to give different penalties. Specifically, the samples staying at the lower margin are given the larger penalties than those in the boundary of the rough margin. Therefore, the new classifier can avoid the over-fitting problem to a certain extent and yields great generalization performance. Experimental results on one artificial dataset and eight benchmark datasets demonstrate the feasibility and validity of our proposed algorithm.