Twin support vector machine with two margin hyperplanes and weighted linear loss

Abstract

Abstract In this paper, we formulate a classification model based on twin support vector machine (TSVM), called twin margin distribution machine with equality constraints (ETMDM). The ETMDM determine two margin hyperplanes by solving two linear equations. The margin hyperplanes are used to replace the boundary hyperplanes in TSVM, resulting in the elimination of inequality constraints. Moreover, the margin hyperplanes investigate the margin distribution information of all samples by the margin mean and margin variance in large margin distribution machine (LDM). And the margin mean and margin variance are reconstructed by weighted linear loss and optimization scheme. The reconstructed margin distribution information can avoid suffering from the possible negative infinity problem and improve the computational efficiency. The experimental results on different types of datasets indicate that our ETMDM has excellent classification accuracy but with less computational time.