KDnuggets : News : 2001 : n01 : item4    (previous | next)

Software

From: Dan Steinberg dstein@salford-systems.com
Date: Tue, 2 Jan 2001 15:15:35 -0800 (PST)
Subject: Salford Systems Announces MARS 2.0 Automated Modeling/Regression Tool
MARS (Multivariate Adaptive Regression Splines) Version 2.0 is
available for immediate evaluation at
http://www.salford-systems.com/evals/marsreg.html.  Also download
http://www.salford-systems.com/MARS.pdf (caution: large PDF, 158 pages)
for the full mathematical foundations.

MARS is one of the core data mining products offered by KDDCup 2000
Winner Salford Systems, and has been successfully deployed in over 200
data mining projects worldwide. Applications have included web site
personalization, response and churn modeling, new drug development,
credit risk assessment, weather forecasting, financial portfolio
adjustment, and a broad range of market research problems.  The
software requires no prior knowledge of MARS and automatically builds
regression and response models.  In a typical data mining application
MARS will explore on the order of 10 to 100 million regression models
to detect the best performing model on test data.  MARS runs typically
require from 5 minutes to 2 hours per model. Core features include:

	Direct access to over 85 different database formats and ODBC
	Automatic variable selection and optimal variable transformation
	Automatic interaction detection
	Automatic missing value handling
	Export of models to C-code, XML, PMML
	Graphical display of key drivers in nay model, 2-D and 3-D
	Automatic report generation with graphs, charts, diagnostics in RTF
		  format for easy publication to Web

Originally developed by Stanford University physicist Jerome Friedman
(also one the authors of CART) and licensed exclusively to Salford
Systems, MARS has proven highly competitive with neural nets while
being much faster and fully understandable.  See our web page for
further information and success stories.

KDnuggets : News : 2001 : n01 : item4    (previous | next)

Copyright © 2001 KDnuggets. Subscribe to KDnuggets News!