Linear Programming Support Vector Regression Method Based On One-class Classification

SVM(Support Vector Machine, referred to as SVM) is an excellent algorithm, which is based on Statistical Learning Theory(Statistical Learning Theory, referred to as SLT). Statistical Machine Learning Theory specifically solve the small sample situation. The traditional statistical methods appear "over-learning" and other issues, which is due to statistical principles of traditional statistical methods is to approximate the sample data as infinite, but the sample data is limited in the actual situation. Therefore Comparing with them, Statistical Learning Theory more adapt to the actual problems in reality. Then SVM based on this theory have a good generalization performance. While the idea of Kernel function is introduced, not only deal with the nonlinear case, more to avoid “dimension disaster” and reduce the actual operation difficulty.SVM is proposed for two-class classification.Due to its good performance, people gradually extended it to other classification and regression problems. At present, the SVC has been relatively perfect, while the SVR is improving. One of SVR study is to improve the algorithm. Because SVR evolved from SVC, they are closely linked. We use two-class classification and regression as an example to prove the equivalence of this theory. Based on this We can apply the existing classification algorithm to the regression,then which become a new regression algorithm. This paper provide a new algorithm by this method.This paper that mentioned newly algorithm is based on one-class classification thoughts,which is extremely rare classification problems,aiming at the detection of abnormal value. So the newly method about algorithm refers programming, this method is superior to the former method: higher arithmetic speed. Through the experiment of sine function, chaotic time series data, we make a comparison between newly algorithm and ?-SVM,LP-SVM and LS-SVM. It concludes that generalization performance of the new algorithm is superior to other methods, which turns out that newly algorithm is effective and available.The newly algorithm provided reflects the relationship between one-class classification and regression, which help us understand the geometric meaning about SVR.It inspires us try to apply more classification algorithm to regression problems.