Online model-based twin parametric-margin support vector machine

Abstract

In this paper, an online model-based TPMSVM (OTPMSVM) is proposed in a reproducing Kernel Hilbert space (RKHS). By exploiting the techniques of the Lagrange dual problem like SVM, the solution of this OTPMSVM can be obtained iteratively and the solution of each iteration can be much efficiently determined. It is illustrated that, based on the experimental results, the accuracy of OTPMSVM is comparable with SVM and other state-of-the-art algorithms. On the other hand, the computational cost of this OTPMSVM algorithm is comparable with or slightly lower than existing online methods, but much lower than that of SVM-based algorithms. © 2014 Springer International Publishing Switzerland.