Robust regression tree-based methods

The paper addresses the problem of robustness of regression trees with respect to outlying values in the dependent variable. New robust tree-based procedures are described which are obtained by introducing in the tree building phase some objective functions already used in the linear robust regression approach, namely the Huber and the Tukey bisquare functions. The performances of the new procedures are evaluated through a Monte Carlo experiment, in which regression trees based on the least squares and the least absolute deviation criteria are also examined.