Sparse and Robust Reject Option Classifier using Successive Linear Programming

  • 2018-02-12 18:38:26
  • Kulin Shah, Naresh Manwani
  • 1

Abstract

In this paper, we propose a new sparse and robust reject option classifierbased on minimization of $l_1$ regularized risk under double ramp loss$L_{dr,\rho}$. We use DC programming to find the risk minimizer. The algorithmsolves a sequence of linear programs to learn the reject option classifier.Moreover, we show that the risk under $L_{dr,\rho}$ is minimized by generalizedBayes classifier in the reject option setting. We also provide the excess riskbound for $L_{dr,\rho}$. We show the effectiveness of the proposed approach byexperimenting it on several real world datasets.

 

Introduction (beta)

None

 

Conclusion (beta)

None