Tree-based Non-linear Graphical Models

Graphical models are statistical models that are associated to graphs whose nodes represent variables of interest. The absence of an edge between two nodes corresponds to a conditional independence between the variables. In this work, I propose a class of graphical models for non-linear systems, where the shape of dependence is modelled by a Bayesian additive regression tree model. The proposed models are able to detect nonparametrically both non-linearities and interactions and are suitable for high dimensional data.