| Popis: |
In this paper, we present new optimization models for Support Vector Machine (SVM), with the aim of separating data points in two or more classes. The classification task is performed by means of nonlinear classifiers induced by kernel functions. Along with a deterministic formulation in which data are assumed to be perfectly known, we propose a robust extension with bounded-by-norm uncertainty sets. Robust optimization techniques allow to protect the models against uncertainties arising during the collection of real-world observations. Closed-form expressions for the bounds of the uncertainty sets in the feature space for typically used kernel functions have been derived. Extensive numerical results on real-world datasets show the benefits of the proposed classifiers in comparison with various alternatives in the SVM literature. |
