A polynomial smooth twin support vector machines based on Newton-Armijo optimization
WEI Xiuxi1 , HUANG Huajuan1，2*
（1 College of Artificial Intelligence, Guangxi University for Nationalities, Nanning 530006, Guangxi, China;2 Key Laboratory of Network Communication Engineering, Guangxi University for Nationalities, Nanning 530006, Guangxi, China）
Abstract:
In order to solve the problem of low approximation accuracy of sigmoid smooth function adopted by smooth twin support vector machines (STWSVM), a polynomial smooth twin support vector machines based on Newton-Armijo optimization (PSTWSVM-NA) is proposed. In PSTWSVM-NA, the positive sign function is introduced to transform two quadratic programming problems of TWSVM into two non-differentiable unconstrained optimization problems. Then, a family of polynomial smooth functions are introduced to smooth the non-differentiable unconstrained optimization problem, and Newton Armijo method with fast convergence speed is used to solve the new model. It is proved theoretically that the PSTWSVM-NA model has any order smoothness. Experimental results on artificial data and UCI data sets show that the algorithm has higher classification accuracy and faster training efficiency.
KeyWords:
twin support vector machines; polynomial; smooth; Newton-Armijo method
