An integer programming approach for the 2-class single-group classification problem

Abstract

Two sets XB, XR ⊆ Rd are linearly separable if their convex hulls are disjoint, implying that a hyperplane separating XB from XR exists. Such a hyperplane provides a method for classifying new points, according to the side of the hyperplane in which the new points lie. In this work we consider a particular case of the 2-class classification problem, which asks to select the maximum number of points from XB and XR in such a way that the selected points are linearly separable. We present an integer programming formulation for this problem, explore valid inequalities for the associated polytope, and develop a cutting plane approach coupled with a lazy-constraints scheme.