Twin Support Vector Machine Local Structural Information for Pattern Classification

Many versions of support vector machine with structural information exploit the useful prior knowledge to directly improve the algorithm's generalization. The prior knowledge embodies the structure of data, but it cannot fully reflect the local nonlinear structure of data. In this paper, a twin support vector machine with local structural information (LSI-TSVM) is proposed. The LSI-TSVM embeds the local within-class and between-class distribution information of data, which makes it contain not only the original global within-class clustering and between-class margin but also the local within-class and between-class scatters. Furthermore, our LSI-TSVM is extended to a nonlinear version with a kernel trick. All experiments show that our LSI-TSVM is superior to the state-of-the-art algorithms in a generalization performance.