ABSTRACT

Regression trees are similar to additive models in that they represent a compromise between the linear model and the completely nonparametric approach. Tree methodology has roots in both the statistics and computer science literature. A precursor to current methodology was CHAID developed by Morgan and Sonquist (1963) although the book by Breiman et al. (1984) introduced the main ideas to statistics. Concurrently, tree methodology was developed in machine learning starting in the 1970s — see Quinlan (1993) for an overview.