半连续Bernoulli-Normal两部回归模型及变量选择

Semi-continuous Bernoulli-Normal Two-part Regression Model and Variable Selection

对于半连续两部回归模型,考虑到每个回归部分都会遇到大量的候选变量,此时就会产生变量选择问题。文章主要研究Bernoulli-Normal两部回归模型的变量选择问题。先提出一种基于Lasso惩罚函数的变量选择方法,但考虑到Lasso估计量不具有Oracle性质,又提出一种基于自适应Lasso惩罚函数的变量选择方法。模拟结果表明:两种方法都能够对Bernoulli-Normal回归模型进行变量选择,且自适应Lasso方法的变量选择性能往往优于Lasso方法。

For semi-continuous two-part regression model,considering that each regression part will encounter a large num⁃ber of candidate variables,a problem of variable selection will arise.This paper mainly studies the variable selection of Bernoul⁃li-Normal two-part regression model.Firstly,a variable selection method based on Lasso penalty function is provided.Then,con⁃sidering that the Lasso estimator does not have the Oracle property,a variable selection method based on adaptive Lasso penalty function is proposed.The simulation results show that both two methods can select variables for Bernoulli-Normal two-part re⁃gression model,and the performance of the adaptive Lasso method is better than the Lasso method.