| Expectile regression model is one of the important statistical methods,which can carefully depict the tail behavior of the response variable.However,classical nonlinear expectile regression has two shortcomings.It is difficult to choose a nonlinear function,and it does not consider the interaction effects among explanatory variables.Therefore,research on a more general nonlinear expectile regression model and to portray nonlinear pattern of the economic and social problems is of great theoretical and practical value to application and dissemination.This paper selects "Expectile Regression Forest Model and Its Application" as the topic of this dissertation.Our proposal is motivated by the proven success of random forest for quantile regression problem.we extend the random forest to expectile regression,the expectile regression forest model and expectile regression forest model based on the BCor-SIS method for high-dimensional data are established respectively,and good numerical simulation and application research results are obtained.The main work and innovation of the paper are as follows:(1)Expectile regression forest model(ERF)is established.The ERF model can not only simulate the nonlinear structure in the real problem well by the random forest model,but also show the complete condition distribution of the explanatory variable to the response variable by expectile regression,which has a wide application prospect in practice.Firstly,the complete modeling steps of the ERF model are given.Secondly,the estimation effect and prediction ability of ERF model are studied by numerical simulation,and compared with additive expectile regression,expectile regression tree,expectile regression gradient boosting tree,expectile regression neural network model,and so on.It is found that the former model is better than the latter in estimation and prediction results.Finally,the ERF model is applied to the real data to verify the validity of the ERF model.(2)Expectile regression forest model based on the BCor-SIS method for highdimensional data(FS-ERF)is established.The FS-ERF model combines screening method of high-dimensional data with expectile regression forest model to solve the problem that Expectile regression forest model runs on high-dimensional data for a long time.Firstly,the complete modeling steps of the FS-ERF model are given.Secondly,through the numerical simulation,it is found that the FS-ERF model is significantly better than the ERF model in the estimation effect,prediction ability and running time,and the results of measure of relative importance are almost the same as the ERF model.Finally,the FS-ERF model was applied to lung cancer genome data to actually test the effectiveness of the model. |
