Bayesian variable selection using an informed reversible jump in imaging genetics: an application to schizophrenia

This is a summary result of my master’s thesis. In this work, we propose a data driven reversible jump algorithm for variable selection. The main idea is that, the next covariate to be included is chosen with probability proportional to the correlation between the candidate covariate and the residuals of the current model. While for deletion, the probability of deletion is inversely proportional to the size of its coefficient.