SECRET (Scalable EM Classification-based Regression Trees) is a new construction algorithm for regression trees with linear models in the leaves. The algorithm produces regression trees with accuracy comparable to existing algorithms and at the same time requires far less computational effort on large datasets. Experimental results show that SECRET improves the running time of regression tree construction by up to two orders of magnitude compared to previous work while constructing trees of comparable quality.
The SourceForge download page has instructions on downloading the initial alpha release of the code.
CVS access is also available:
Please send email to or contact the authors directly:
Alin Dobra, Johannes Gehrke.
SECRET: A Scalable Linear Regression Tree Algorithm.
In Proceedings of the Eighth
ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.
Edmonton, Alberta, Canada, July 2002.