🤖 AI Summary
To address the insufficient robustness of multiclass classification arising from feature uncertainty, this paper proposes the Robust Two-Parameter Margin Support Vector Machine (TPMSVM). TPMSVM is the first to integrate robust optimization into the two-parameter margin SVM framework: it models input perturbations via a norm-bounded uncertainty set and derives a tractable convex quadratic programming formulation. The method supports both linear and kernel-based multiclass classification and introduces two theoretically sound, discriminative decision functions. Extensive experiments on multiple real-world benchmark datasets demonstrate that TPMSVM significantly outperforms conventional multiclass SVMs and state-of-the-art twin SVM variants in classification accuracy, noise resilience, and generalization stability. The approach thus achieves a rigorous theoretical foundation while maintaining strong practical effectiveness.
📝 Abstract
In this paper, we present novel Twin Parametric Margin Support Vector Machine (TPMSVM) models to tackle the problem of multiclass classification. We explore the cases of linear and nonlinear classifiers and propose two possible alternatives for the final decision function. Since real-world observations are plagued by measurement errors and noise, data uncertainties need to be considered in the optimization models. For this reason, we construct bounded-by-norm uncertainty sets around each sample and derive the robust counterpart of deterministic models by means of robust optimization techniques. Finally, we test the proposed TPMSVM methodology on real-world datasets, showing the good performance of the approach.
