| Count data is widespread in the fields of biomedicine,industry,agriculture and so on.To process this kind of data,we usually use some discrete models,such as Poisson model and negative binomial model.However,this type of count data often contains a large number of zeros and the standard discrete distribution is no longer suitable for them,so zero-inflated model which caused widespread concern in recent years has become an effective method to analyze this type of data.In addition,a large number of variables are often involved in the actual data.In order to establish a reasonable model,it is necessary to select variables.In this paper,the problem of variable selection is discussed with the zero-inflated Poisson regression model.This paper firstly introduces the zero-inflated Poisson model.Then,based on lasso,elastic net and SCAD penalty functions,the corresponding logarithmic likeli-hood of penalty for the zero-inflated regression model is given.And then,we build pseudo-data based on Taylor approximation algorithm,and use coordinate descent method to studies the variable selection methods.Then under the penalty of lasso and elastic net,the paper based on Gibbs sampling and MH algorithm studies the variable selection methods such as Bayes lasso and Bayes elastic net.In order to demonstrate the effectiveness of the proposed methods,the simulation studies of variable selection under different sample sizes,different zero proportions and different punishments are presented.Finally,the zero-inflated Poisson model is used to select variables from a group of outpatient data of hospitals to furtherd demonstrate the effectiveness of the proposed methods in this paper. |
