The SVR Algorithm Under Different Loss Function And Its Application

This thesis learns to understand SVM from the angle of statistical.Compares SVR algorithms under different loss function for further studies.The thesis follows these steps:Chapter 1 introduces the importance of solving classification problems and regression problems and the advantages to solve these problems with SVM.The basic methods and algorithms about SVC and SVR are summarized in the second chapter.Chapter 3 introduces how to understand SVM from the angle of statistics.With the concept of loss function,it's nature to combine SVC and SVR as the same kind of problem-to minimize the empirical risk.In the case of right loss function,deduced the algorithm of SVC and SVR.In a certain circumstance of knowing conclusions,a supplemental proof about SVR algorithm of general loss function is derived with the duality theory.SVR algo-rithms of four well-known loss functions are concrete with application of the proof.Try to construct loss function and give concrete SVR expression with the conclusions above.After all,also give the reader some train of thoughts to construct loss function himself.In chapter 4,the simulation experiment contrast is done between different loss functions with the data of(?)plus random disturbance.Judge the effect of different loss functions and find the cause of Judge the effect of different loss function and find the cause of effect different effects.Give the proof that the construction of loss function is successful in the case of the MSE.Chapter 5 gives the conclusion.