A New Support Vector Regression Model

Support Vector Machines (SVM) based on statistical learning theory is a intelligent machine learning that developed by Vapnik in the middle of the 1970s, it contains Support Vector Clas-sification and Regression,and as we known until 1990s.Vapnik propose some new opinions such as the "structure risk minimization" with optimization. As the maturity of theory, SVM can successfully solve many difficult machine learning problems, such as non-linearity, small sample size, high-dimension of feature space.Comparing with the classification of SVM, the researches of regression has many challenge work to solve.In this paper, we propose a new support vector regression machine named MTRM(Minimum Tube Regression Machine) utterly advoids the parameter redundancy by the technology which transform theε-tube hard linerity regression progrem to a quadratic programming, and to deal with two defect that one is the parameter redundancy inε-SVR and the other is lossing the "support vector sparse". MTRM utilizes the penalty function and kernel function to transform the hard linerly regression to nonlinerly regerssion model. The numerial result shows MTRM doesn't increase the model complexity, and remain have the advantage of support vector sparse while advoid the parameterε; The result is the same between MTRM andε- SVR wtih the optimal parameterε.