GPBoost for Grouped and Areal Spatial

 

Mixed Effects Machine Learning with GPBoost for Grouped and Areal Spatial Econometric Data

The GPBoost algorithm extends linear mixed effects and Gaussian process models by replacing the linear fixed effects function with a non-parametric non-linear function modeled using tree-boosting. This article shows how the GPBoost algorithm implemented in the GPBoost library can be used for modeling data with a spatial and grouped structure. We demonstrate the functionality of the GPBoost library using European GDP data which is an example of areal spatial econometric data.

Applying a GPBoost model (= combined tree-boosting and random effects / GP models) involves the following main steps:

Define a GPModel in which you specify the following:

- A random effects model (e.g., spatial random effects, grouped random effects, combined spatial and grouped, etc.)

- The likelihood (= distribution of the response variable conditional on fixed and random effects)

Create a gpb.Dataset with the response variable and fixed effects predictor variables

Choose tuning parameters, e.g., using…

Here