Bayesian additive regression trees with model trees

Bayesian Additive Regression Trees (BART) is a tree-based machine learning method that has been successfully applied to regression and classification problems. BART assumes regularisation priors on set of trees work as weak learners very flexible for predicting in the presence non-linearity high-order interactions. In this paper, we introduce an extension BART, called Model (MOTR-BART), considers piecewise linear functions at node levels instead constants. MOTR-BART, rather than having unique value level prediction, predictor estimated considering covariates have used split variables corresponding tree. our approach, local linearities are captured more efficiently fewer required achieve equal or better performance BART. Via simulation studies real data applications, compare MOTR-BART its main competitors. R code implementation available https://github.com/ebprado/MOTR-BART.