Enhanced algorithm for high-dimensional data classification

Minimum class variance support vector machine (MCVSVM) and large margin linear projection (LMLP) classifier, in contrast with traditional support vector machine (SVM), take the distribution information of the data into consideration and can obtain better performance. However, in the case of the singularity of the within-class scatter matrix, both MCVSVM and LMLP only exploit the discriminant information in a single subspace of the within-class scatter matrix and discard the discriminant information in the other subspace. In this paper, a so-called twin-space support vector machine (TSSVM) algorithm is proposed to deal with the high-dimensional data classification task where the within-class scatter matrix is singular. TSSVM is rooted in both the non-null space and the null space of the within-class scatter matrix, takes full advantage of the discriminant information in the two subspaces, and so can achieve better classification accuracy. In the paper, we first discuss the linear case of TSSVM, and then develop the nonlinear TSSVM. Experimental results on real datasets validate the effectiveness of TSSVM and indicate its superior performance over MCVSVM and LMLP.