1-NRTSVM via ADMM for Automatical Feature Selection and Classification Simultaneously

摘要

In this paper, a novel classifier with linear and nonlinear versions for data classification and automatical feature selection simultaneously is proposed and named as 1-norm regularized twin support vector machine (1-NRTSVM). By means of the alternating direction method of multipliers (ADMM), two implementation algorithms for 1-NRTSVM are presented. A major feature of the proposed method is directly solving primal problems not dual problems. Experiment results show that the proposed 1-NRTSVM is an effective and competitive classifier for data classification and automatical feature selection.