In this paper we present a Twin Parametric-Margin Support Vector Machine (TPMSVM) model to tackle the problem of multiclass classification. In the spirit of one-versus-all paradigm, for each class we construct a classifier by solving a TPMSVM-type model. Once all classifiers have been determined, they are combined into an aggregate decision function. We consider the cases of both linear and nonlinear kernel-induced classifiers. In addition, we robustify the proposed approach through robust optimization techniques. Indeed, in real-world applications observations are subject to measurement errors and noise, affecting the quality of the solutions. Consequently, data uncertainties need to be included within the model in order to prevent low accuracies in the classification process. Preliminary computational experiments on real-world datasets show the good performance of the proposed approach.

Robust Twin Parametric Margin Support Vector Machine for Multiclass Classification