Spline function smooth support vector machine for classification
Yubo Yuan, Weiguo Fan, Dong-Mei Pu · Journal of Industrial and Management Optimization · 2007
Support vector machine (SVM) is a very popular method for binarydata classification in data mining (machine learning). Since theobjective function of the unconstrained SVM model is a non-smoothfunction, a lot of good optimal algorithms can't be used to findthe solution. In order to overcome this model's non-smoothproperty, Lee and Mangasarian proposed smooth support vectormachine (SSVM) in 2001. Later, Yuan et al. proposed the polynomialsmooth support vector machine (PSSVM) in 2005. In this paper, athree-order spline function is used to smooth the objectivefunction and a three-order spline smooth support vector machinemodel (TSSVM) is obtained. By analyzing the performance of thesmooth function, the smooth precision has been improved obviously.Moreover, BFGS and Newton-Armijo algorithms are used to solve theTSSVM model. Our experimental results prove that the TSSVM modelhas better classification performance than other competitivebaselines.