A0468
Title: Loss-based prior for tree topologies in BART models
Authors: Fabrizio Leisen - Kings College London (United Kingdom) [presenting]
Cristiano Villa - Duke Kunshan University (China)
Kevin Wilson - Newcastle University (United Kingdom)
Francesco Serafini - University of Bristol (United Kingdom)
Abstract: The purpose is to present a novel prior for tree topology within Bayesian additive regression trees (BART) models. This approach quantifies the hypothetical loss in information and the loss due to the complexity associated with choosing the wrong tree structure. The resulting prior distribution is compellingly geared toward sparsity, a critical feature considering BART models' tendency to overfit. The method incorporates prior knowledge into the distribution via two parameters that govern the tree's depth and balance between its left and right branches. Additionally, a default calibration is proposed for these parameters, offering an objective version of the prior. The method's efficacy is demonstrated on both simulated and real datasets.
