A Sparse Support Vector Machine Classifier with Nonparametric Discriminants

Abstract

This paper introduces a novel Sparse Support Vector Machine model with Kernel Nonparametric Discriminants (SSVMKND) which combines data distribution information from two classifiers, namely, the Kernel Support Vector Machine (KSVM) and the Kernel Nonparametric Discriminant (KND). It is a convex quadratic optimization problem with one global solution, so it can be estimated efficiently with the help of numerical methods. It can also be reduced to the classical KSVM model, and existing SVM programs can be used for easy implementation. We show that our method provides a sparse solution through the Bayesian interpretation. This sparsity can be used by existing sparse classification algorithms to obtain better computational efficiency. The experimental results on real-world datasets and face recognition applications show that the proposed SSVMKND model improves the classification accuracy over other classifiers and also provides sparser solution.