Keywords: Zellner’s g-priors, Bayesian Regression, Subset Selection, Genetic Algorithm, Entopic Complexity, and ICOMP.
This e-poster introduces and develops Bayesian regression models under a class of celebrated Zellner’s g-priors for subset selection of best predictors using the genetic algorithm (GA) and information complexity (ICOMP) criterion as our fitness function. We introduce a new entropic complexity based g-priors to score in addition to the existing g-priors in the literature for subset selection. We perform a large scale Monte Carlo simulation study under different g-priors in Bayesian regression models to choose the best g-prior among the candidate g-priors including the entropic complexity based g-priors in subset selection of variables. Our results show the performance of each g-prior. We also show our results on a benchmark data set to demonstrate the utility and versatility of our proposed approach using the GA and ICOMP as the fitness function. The results obtained show that entropic complexity based g-priors outperform currently used existing g-priors in the Bayesian regression models.