Variable Selection Of High-dimensional Mixture Model

Mixture experimental design is a kind of design which assumed that the re-sponse variables only depend on the components in the mixture and not on the amount of the mixture.What we have always been concerned about is selecting the important components from many components of mixtures in order to reduce the number of design points.Subset selection is suitable for low-dimensional mixture model,but when the mixture components increase,the complexity of the algorithm increases and the result of variable selection will change greatly with the slight change of data.In view of these shortcomings,this paper studies the variable selection problem of high-dimensional mixture model.Firstly,in this paper,the application of shrinkage methods such as LASSO,Adaptive LASSO,Elastic net and SCAD to high-dimensional mixture model can screen out important variables quickly,get point estimates of parameters and reduce the number of design points.It is proved that SCAD is better than other three methods through real data analysis.Secondly,we study the variable selection of high-dimensional mixture model using Bayesian method.Compared to traditional LASSO method,Bayesian LAS-SO can not only select variables and estimate parameters,but also provide more reliable interval estimates of parameters,which are illustrated by examples.Finally,we summarize the main contents of this paper and introduce the future work.