Tree models: a Bayesian perspective

Abstract

Classical tree models represent an attempt to create nonparametric models which have good predictive powers as well a simple structure readily comprehensible by non- experts. Bayesian tree models have been created by a team consisting of Chipman, George and McCulloch and second team consisting of Denison, Mallick and Smith. Both approaches employ Green's Reversible Jump Markov Chain Monte Carlo tech- nique to carry out a more e®ective search than the `greedy' methods used classically. The aim of this work is to evaluate both types of Bayesian tree models from a Bayesian perspective and compare them.