Comparing Simplification Methods for Model Trees with Regression and Splitting Nodes

In this paper we tackle the problem of simplifying tree-based regression models, called model trees, which are characterized by two types of internal nodes, namely regression nodes and splitting nodes. We propose two methods which are based on two distinct simplification operators, namely pruning and grafting. Theoretical properties of the methods are reported and the effect of the simplification on several data sets is empirically investigated. Results are in favor of simplified trees in most cases.