Software

Salford Systems is releasing the software used in its recent win in the KDDCup 2000 web mining competition. Key features of CART 4.0 instrumental in supporting the winning entry include

(1) capability of handling nominal predictors with an unlimited number of levels ( thousands of referring URLs appeared in the KDDCup data),

(2) application of penalties to predictors to reflect percent missing in the data, number of levels in a nominal variable, or the cost of data acquisition,

(3) creation of surrogate splitters at every node to assist in interpetating the tree,

(4) built-in bagging and ARCing (boosting variant),

(5) on-screen mouse-controlled pruning and

(6) reports displaying training and test data results at the level of the entire tree, at the level of any subtree rooted anywhere, or within a node.

The CART 4.0 report writer summarizes individual trees and entire sessions as well as collections of trees and supports easy publication to the Web. CART 4.0, which is robustly capable of handling files with over 8,000 predictors and millions of records at high speed, is in widespread use in Fortune 1000 data mining projects.

Further information on CART 4.0, individual, group, and enterprise wide licensing, and Salford Systems data mining and web mining services can be obtained at http://www.salford-systems.com.