DOI: 10.21276/sjet.2017.5.2.4

Pages: 54-57

The structure of K-SVCR algorithm is „one-against-one-against-rest‟． Its advantage is in the process of each decomposition , make all training points of information have been fully taken advantage of. To a certain extent, can prevent the classification error by incomplete information . But this algorithm constructed a quadratic programming which restricted this algorithm‟s speed and the range of applications. So, this paper constructed a linear programming model based on K-SVCR , and then adopt the effective algorithm of Predictor-corrector Method of Mehrotra to solve the linear programming. Preliminary numerical experiments on benchmark datasets show that the algorithm has good performance on both accuracy and training speed than K-class Support Vector Classification-Regression.