Abstract

Bayesian Decision Trees are known for their probabilistic interpretability.However, their construction can sometimes be costly. In this article we presenta general Bayesian Decision Tree algorithm applicable to both regression andclassification problems. The algorithm does not apply Markov Chain Monte Carloand does not require a pruning step. While it is possible to construct aweighted probability tree space we find that one particular tree, thegreedy-modal tree (GMT), explains most of the information contained in thenumerical examples. This approach seems to perform similarly to Random Forests.

 

Quick Read (beta)