The soft-margin Support Vector Machine with ordered weighted average

This paper deals with a cost sensitive extension of the standard Support Vector Machine (SVM) using an ordered weighted sum deviations misclassified individuals respect to their corresponding supporting hyperplanes. In contrast previous heuristic approaches, exact method that applies average operator in classical SVM model is proposed. Specifically, when weights are sorted non-decreasing order, quadratic continuous formulation developed. For general weights, mixed integer addition, our results prove nonlinear kernel functions can be also applied these new models extending its applicability beyond linear case. Extensive computational reported show predictive performance provided by proposed solution approaches better than ones (linear and kernel) similar or Maldonado et al. (2018).