An explicit split point procedure in model-based trees allowing for a quick fitting of GLM trees and GLM forests

Classification and regression trees (CART) prove to be a true alternative full parametric models such as linear (LM) generalized (GLM). Although CART suffer from biased variable selection issue, they are commonly applied various topics used for tree ensembles random forests because of their simplicity computation speed. Conditional inference model-based algorithms which is tackled via fluctuation tests known give more accurate interpretable results than CART, but yield longer times. Using closed-form maximum likelihood estimator GLM, this paper proposes split point procedure based on the explicit in order save time when searching best given splitting variable. A simulation study non-Gaussian response performed assess computational gain building GLM trees. We also propose benchmark simulated empirical datasets against conditional LM identify situations where efficient. This approach extended multiway log-transformed distributions. Making possible through new allows us investigate use ensemble methods. numerical comparison other forest-type approaches. Our analyses show cases good challengers forests.