Computational Enhancements in Tree-Growing Methods
In this paper we show how to avoid unnecessary calculations and to save considerably the computational cost in a wide class of tree-based methods. So called auxiliary statistics, which enable to restrict handling the raw data, are introduced. Aside that, a fast splitting algorithm is outlined, which allows to recognize and avoid unnecessary split evaluations during the search of an optimal split. Relationships between the computational cost savings and properties of both a specific method and data are summarized.
Key WordsClassification and regression trees computational cost auxiliary statistics fast splitting algorithm
Unable to display preview. Download preview PDF.
- Antoch J. and Ekblom H. (1995). Recursive robust regression, computational aspects and comparison, CSDA, 19, 115–128.Google Scholar
- Breiman L., Friedman J.H., Olshen R.A. and Stone C.J. (1984). Classification and Regression Trees, Wadsworth, Belmont CA.Google Scholar
- Ciampi A. (1993). Constructing prediction trees from data, the RECPAM approach, in: Computational Aspects of Model Choice, Antoch J. (Ed.), Physica - Verlag, Heidelberg, 105–156.Google Scholar
- Klaschka J. and Antoch J. (1997). How to grow trees fast, in: ROBUST’96, Antoch J. and Dohnal G. (Eds.), JČMF, Prague, 91–106 (in Czech).Google Scholar
- Mola F. and Siciliano R. (1998). A general splitting criterion for classification trees, Metron, to appear.Google Scholar
- Siciliano R. and Mola F. (1996). A fast regression tree procedure, in: Statistical Modelling, Proc. 11th Int. Workshop on Statistical Modelling, Forcina A. et al. (Eds.), Graphels, Perugia, 332–340.Google Scholar