Partially Fixed Bayes Additive Regression Trees for spatial-temporal related model

TL;DR

This paper introduces a partially fixed BART model tailored for spatial-temporal data, enhancing estimation accuracy and efficiency by incorporating prior structural information, demonstrated through data experiments and real-world examples.

Contribution

The paper proposes a novel partially fixed BART model that improves estimation accuracy and efficiency for spatial-temporal data by integrating prior structural information.

Findings

Improved estimation accuracy over original BART.

Enhanced efficiency in model fitting.

Better structural insights for future analysis.

Abstract

Bayes additive regression trees(BART) is a nonparametric regression model which has gained wide -spread popularity in recent years due to its flexibility and high accuracy of estimation .In spatio-temporal related model,the spatio or temporal variables are playing an important role in the model.The BART models select variables with uniform prior distribution that means treat every variable equally.Applying the BART model directly without properly using these prior information is not appropriate.This paper is aimed at a modification to the BART by fixing part of the tree's structure.We call this model partially fixed BART.By this new model we can improve efficiency of estimation.When we don't know the prior information,we can still use the new model to get more accurate estimation and more structure information for future use.Data experiments and real data examples show the improvement comparing to the original Bart model.