International Journal of Applied Mathematics and Machine LearningAuthor's: Xinxin Zhang and Liya Fan
Pages: [27] - [45]
Received Date: February 7, 2015
Submitted by: Jianqiang Gao.
This paper is devoted to extend projective twin support vector machine (PTSVM) by smoothing technique and propose a navel classification method with linear and nonlinear versions, named as smoothed projective twin support vector machine (SPTSVM). The advantage of SPTSVM is to solve a pair of unconstraint differentiable optimization problems rather than a pair of dual QPPs. By means of Newton-Armijo method, an effective fast algorithm is suggested for solving SPTSVM. Experiment results compared with SVM, TSVM, PTSVM, and SPTSVM show that the proposed SPTSVM is a fast and effective classification method.
projective twin support vector machine, plus function, smoothing approximation function, unconstraint differentiable optimization, Newton-Armijo method.
