An alternative robust local embding based on twin support vector machines

Aiming at the problem that many existing nonparallel hyperplane support vector machine (NHSVM) methods only considered the global information of the training samples in the same class and did not fully take into account the local geometric structure and the underlying descriminant information, an alternative robust local embedding based twin support vector machine (ARLEBTSVM) was presented by introducing the basic theories of alternative robust local embedding (ARLE) algorithm into the NHSVM. ARLEBTSVM not only inherits the characteristic of NHSVM methods which can well deal with the XOR problem, but also fully considers the local and global geometric structure of training samples in the same class and shows the local and global underlying discriminant information. In addition, in order to well deal with the nonlinear classification problem, the kernel mapping method was used to extend ARLEBTSVM to the nonlinear case. Experimental results on some artificial datasets and many real UCI datasets indicate that the proposed ARLEBTSVM method has better classification ability. ©, 2015, Central South University of Technology. All right reserved.