A Robust Optimization Model for Nonlinear Support Vector Machine
From MaRDI portal
Abstract: In this paper we present new optimization models for Support Vector Machine (SVM), with the aim of separating data points in two classes by means of a nonlinear classifier. Traditionally, in the nonlinear context data points are firstly mapped to a higher dimensional space and then classified through a SVM-type model. In order to increase the predictive power of SVM, within our approach we include a final linear search procedure aiming to minimize the overall number of misclassified points. Along with a deterministic model in which data are assumed to be perfectly known, we formulate a robust optimization model with bounded-by-lp-norm uncertainty sets. Indeed, when data are real-world observations, measurement errors or noise may corrupt the quality of input values. For this reason, facing uncertainty in the model is a way to robustify the approach. All formulations reduce to linear models with advantages in terms of efficiency compared to other approaches in the literature. Extensive numerical results on real-world datasets show the benefits in terms of accuracy when considering nonlinear decision classifier and protecting the model against uncertainties. Finally, managerial insights to guide the final user are provided.
This page was built for publication: A Robust Optimization Model for Nonlinear Support Vector Machine
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6439844)
