Abstract:
|
In generalized linear model(GLM), variable selection is the problem of interest, especially in clinical statistical research because of high dimension issue. Bayesian methods has many advantages in solving variable selection problem, most existing Bayeisan methods in GLM employ a prior which put a great mass on density at null value. While nonlocal prior could also efficiently eliminate unnecessary covariates, this theory has been developed in linear model. My contribution is extending this nonlocal prior Bayesian method to GLM, also convergence rate is developed under some conditions. Application result showed this method could recognize the model with good prediction performance, and model result is simpler than LASSO in GLM.
|