Software

RT is a learning system that can handle multiple regression problems
based on samples of data. The resulting models have the form of a
regression tree. Linux and SunOS versions of RT 4.1.0 are now available
for free download for research purposes at:
http://www.liacc.up.pt/~ltorgo/RT/

Compared to existing systems for obtaining regression trees, the most
distinguishing feature of RT is its ability to use local models in the
tree leaves, leading to what we have named local regression trees. The
use of these models brings highly significant gains in terms of
prediction accuracy at the cost of some interpretability of the models
and also with an increase of computation time.

Due to the schema used to integrate local models with regression trees,
RT can easily behave like a normal local modeling system or a standard
regression tree learner. Moreover, this schema allows RT to emulate
other regression modeling techniques (like for instance CART, Core, or
RETIS).

Further information can be obtained at RT web page
(www.liacc.up.pt/~ltorgo/RT/).

Luis Torgo
LIACC/FEP, University of Porto
ltorgo@liacc.up.pt - www.liacc.up.pt/~ltorgo