Somak Dutta
Department of Statistics
Iowa State University

A Bayesian screening method in ultra-high dimensional settings

We propose a novel Bayesian screening method in ultra-high dimensional settings, where the sample size is much smaller than the number of features. The proposed method opens up the possibility of considering prior information on the effect sizes and model size during variable screening. The proposed method iteratively considers the inclusion probabilities of potential variables integrating out the effects of already included variables and includes the variables using the Bayes factor or the BIC. We demonstrate the effect of the hyperparameters on screening properties of the method and scalability of the proposed method using simulation and real data applications. This is a joint work with Vivekananda Roy.

Refreshments at 3:45pm in Snedecor 2101.