A TreeNet® model normally consists of from several dozen to several hundred small trees, each typically no larger than two to eight terminal nodes. The model is similar in spirit to a long series expansion (such as a Fourier or Taylor's series) - a sum of factors that becomes progressively more accurate as the expansion continues. The expansion can be written as:
Contact Salford Systems at 619.543.8880 or e-mail support (at) salford-systems (dot) com. We maintain a collection of white papers and academic studies on various data mining topics on the web site and offer tutorials on TreeNet®, CART®, and MARS® in major cities world wide. Internet meetings to demonstrate and discuss any of our products can be arranged.
The Salford Systems data mining solution rests on two groups of technologies: CART, MARS, and PRIM for accurate, easy-to-understand models, and TreeNet® and RandomForests® for ultra-high performance, potentially complex models interpreted via supporting graphical displays. Even in circumstances where interpretability and transparency are mandatory and a model must be expressed in the form of rules, TreeNet can serve a useful function by benchmarking the maximum achievable accuracy against which interpretable models can be compared.
TreeNet® is designed for very high accuracy predictive modeling. Because TreeNet® attempts to achieve this goal even if very complex models are required, models may be relatively difficult to understand in detail. However, the graphs produced by TreeNet® software display the impact of any relevant predictor or pair of predictors on the target, thus revealing the underlying data structure.
TreeNet® requires that both training and test data reside in RAM. Thus, if large databases are being analyzed, TreeNet® will be most effective when running on large-capacity servers. We recommend a minimum of 512 MB RAM and on Windows machines, Windows XP or later versions of the OS are preferred platforms for performance. TreeNet® is available for Windows XP or later and UNIX (IBM AIX, Compaq Alpha, SGI, HP, and Sun) platforms and will run with as little as 64 MB RAM. A Linux version is planned.
TreeNet® was developed in 1997 by Stanford University's Jerome Friedman, one of the authors of CART®, the author of MARS®, and the inventor of Projection Pursuit and HotSpotDetector®. The TreeNet® technology has been tested in a broad range of industrial and research settings and has demonstrated considerable benefits. In tests in which TreeNet® was pitted against expert modeling teams using a variety of standard data mining tools, TreeNet® was able to deliver results within a few hours comparable to or better than results requiring months of hands-on development by expert data mining teams.