Redefining support vector machines with the ordered weighted average

Abstract

In this work, the classical soft-margin Support Vector Machine (SVM) formulation is redefined with the inclusion of an Ordered Weighted Averaging (OWA) operator. In particular, the hinge loss function is rewritten as a weighted sum of the slack variables to guarantee adequate model fit. The proposed two-step approach trains a soft-margin SVM first to obtain the slack variables, which are then used to induce the order for the OWA operator in a second SVM training. Originally developed as a linear method, our proposal extends it to nonlinear classification thanks to the use of Kernel functions. Experimental results show that the proposed method achieved the best overall performance compared with standard SVM and other well-known data mining methods in terms of predictive performance.