Research On Non-Local Prior Model Based On Lasso Estimator

The rapid development of computing technology has strengthened the ability to acquire data,especially the production of various high-dimensional data.One type of high-dimensional data is characterized by a large number of variables and a small sample size.It appears in various areas of production and life.How to extract effective information is the focus of high-dimensional data applications.This paper describes three major types of methods involved in the selection of high-dimensional data variables:Lasso,Bayesian,and Bayesian regularization methods.This paper replaces the original MAP estimator of non-local prior methods with Lasso estimator.Constructing a new method—non-local prior method based on Lasso estimator to study the effectiveness of the new method.In the Bayesian regularization method,the Bayesian Lasso method is selected and research this method in a high-dimensional environment.Through simulation experiments,Lasso,Adaptive Lasso,Elastic Net,ISIS-SCAD,non-local prior method,Bayesian Lasso,and non-local prior method based on Lasso estimator are used to select variables for randomly generated data sets.Using ATP,C Indicators,ENE indicators to compare the effectiveness of each method in variable selection.The simulation results show that the Lasso method will select more variables and result in a higher ATP,while the non-local prior method will select fewer but more accurate variables,so it will perform better on the C.The non-local prior method based on the Lasso estimator is not better than the non-local prior method,but as the dimensions increase,the non-local prior method based on the Lasso estimator guarantees similar ATP values to the Lasso-like methods while its ENE value is lower than most Lasso methods,so it is better than Lasso methods.The Bayesian Lasso method performs better in low-dimensional situations,but does not show its advantages in high-dimensional situations.For the four data sets,we use non-local prior methods,Lasso and non-local prior methods based on Lasso estimators.The empirical results show that the non-local prior method based on the Lasso estimator can avoid the problem of selecting very few variables by the non-local prior method to a certain extent.And it will not choose more variables like Lasso.