Trees-Based Models for Correlated Data

This paper presents a new approach for treesbased In this paper we develop a method which combines the regression, such as simple regression tree, concepts of random effects and random fields -- which are random forest and gradient boosting, in settings convenient platforms for analyzing correlated data -- and involving correlated data. We show the problems trees-based models such as: regression tree, random forest that arise when implementing standard treesbased and gradient boosting. The desired result is that the treesbased regression models, which ignore the correlation part results a high prediction accuracy and model structure. Our new approach explicitly selection capabilities and the random effects aspect enables takes the correlation structure into account in the to boost the model performance by utilizing correctly the splitting criterion, stopping rules and fitted values correlation structure and even allows statistical inference.