Research On Bagging Method Based On Spline Model

Smoothing spline is one of the most important methods in nonparametric models,construct a penalty term that describes the degree of bending of the function during the fitting of the model,by minimizing the sum of the residuals of the fitted models and the sum of the penalty terms,to get the smooth spline model,And in this process,add a coefficient to the penalty item-smooth parameter to control model for the smooth degree of the degree of punishent.The key step of fitting the smooth spline model is the selection of smooth parameters,its essence is a process of model selection.Many scholars have made research on the uncertainty of model selection,It is pointed out that if we only fit a model for a set of samples,and then the data is analyzed,it will bring more risks to the analysis,and at the same time point out the risk,it shows that model combination is a good way to reduce this uncertainty.The process of selecting smooth parameters when applying the spline model is the process of model selection,which will also bring uncertainty to the application of the spline model.This method combines smooth spline model and bagging method to reduce the smooth parameter uncertainty,in order to improve the accuracy of prediction model,by using stochastic simulation method prove that the proposed method can improve the natural cubic spline model and two-dimensional thin plate spline model forecast accuracy,namely:the natural cubic spline of Bagging based on cross validation method can improve the accuracy of traditional forecasting based on the natural cubic spline cross validation,generalized cross validation,two-dimensional thin plate model Bagging method based on generalized cross validation can improve the prediction accuracy of traditional two-dimensional thin plate model based on generalized cross validation.And on the basis of existing scholars' research,two actual datasets have been selected:data set of blood glucose and insulin determination in patients and The 3D Road Network of the UCI data set,on the two data sets,the method is studied in this paper.The fitting effect of the model is depicted with the mean square error of the model on the test set,on the two dataset,the mean square error of the proposed method is reduced by 26%and 10%respectively.The results of the empirical study also prove the effectiveness of the proposed method.