An improved non-parallel Universum support vector machine and its safe sample screening rule

Abstract

A novel non-parallel hyperplane Universum support vector machine (U-NHSVM) is proposed in this paper. Universum data with ensconced prior knowledge are exploited by a non-parallel hyperplane support vector machine. In contrast to other algorithms, the proposed U-NHSVM shows flexibility by exploiting the prior knowledge ensconced in Universum and provides consistency by constructing two non-parallel hyperplanes simultaneously. With Universum, U-NHSVM is clearly effective but also time consuming. Therefore, a safe sample screening rule (SSSR) for U-NHSVM is also proposed based on its sparsity, termed SSSR-U-NHSVM. Because only the non-SVs are excluded from both labelled and Universum samples, the efficiency of SSSR-U-NHSVM is extremely improved while the accuracy is completely conserved. Numerical experiments on seventeen benchmark datasets and a Chinese wine dataset are carried out to demonstrate the validity of the proposed U-NHSVM and SSSR-U-NHSVM.