THE NEXT TECHNOLOGY ADVANCE IN DATA MINING HAS ARRIVED: SALFORD SYSTEMS LAUNCHES EXCLUSIVE PREDICTIVE MODELING SOLUTION
Salford Systems has introduced MARS, a high-speed flexible regression tool for data mining and predictive modeling.
A natural alternative to Neural Networks and more conventional regression models, MARS was designed to solve problems such as how to predict credit card holder balances, insurance claim losses, and cell phone usage with superior forecasting accuracy. MARS builds its models automatically, self-tests to ensure validity, and graphically displays the effect of each important variable on the outcome.
"MARS is the ideal modeling tool when an analyst needs to accurately predict a future outcome and also needs to understand the 'why' underlying the predictive model," said Dan Steinberg, president of Salford Systems. "Prior to the advent of MARS, data miners had to choose between tools that optimized predictive accuracy or tools that told a story about the underlying data patterns and relationships. Now, with MARS, an analyst no longer has to sacrifice intelligibility to obtain predictive accuracy. Further, the time-consuming trial and error process of building accurate regression models is a thing of the past with the availability of MARS' highly-automated, fast analytical engine and unique model-building capability."
Herb Edelstein, president of Two Crows Data Mining Consultancy and a widely regarded industry analyst, predicts that, "MARS will be one of the next hottest algorithms. MARS addresses some of the shortcomings of decision trees, and it does so in a fairly elegant fashion." Independent studies have demonstrated that MARS not only outperforms neural networks in a wide variety of settings, but is also hundreds of times faster.
The intuitive, Windows-based interface and intelligent default settings enable non-technical business users to run MARS easily. The analyst provides MARS with a database and target variable; MARS then develops a model, self-tests to prevent over-fitting, and graphically displays the impact of each predictive factor on the outcome. Predictions for new data can be obtained either through the MARS engine or via C-source code produced as part of the modeling process.
MARS is ideal for predictive modeling problems involving continuous outcomes, such as how much a customer will spend on a catalog order, the minutes a customer will use his cell phone, or how electricity production will change as generator inputs change. MARS is also adept at predicting "yes/no" binary responses such as whether a consumer will default on a loan, refinance a mortgage, or defect to a competitor.
Experienced data analysts also can use MARS as an exploratory tool to refine and improve more conventional linear and logistic regression models. By automatically detecting needed variable transformations and interactions, MARS can radically reduce the time required to build a model and significantly improve its predictive accuracy.
"MARS is an essential tool for any data miner," said Thomas Brauch, marketing manager of the Data-Driven Marketing Department at Fireman's Fund Insurance. "It finds significant effects in complex data structures where other methods simply fail. I use it as both a stand alone solution and as a transformation tool for simpler modeling techniques."
MARS, an acronym for Multivariate Adaptive Regression Splines, is a non-parametric regression procedure. It extends technology originally introduced in CART, the industry's leading decision-tree software. MARS automates all aspects of model development and deployment, including:
MARS enables analysts to rapidly search through a large number of candidate models and quickly identify the "optimal" solution - providing valuable business understanding in the process. Predictive models can be deployed directly from within the software or exported as ready-to-run C and SAS compatible source code.
MARS is supported in a variety of stand alone and client/server operating environments. Operating systems supported include Windows 95/98, Windows NT, UNIX, and IBM MVS and CMS. Hardware platforms supported include Intel PCs, Sun, SGI, HP, Compaq Alpha, IBM RS6000, and IBM mainframes. MARS' scalability is limited only by a system's available RAM. The software requires a minimum of 10 MB of disk storage and 64 megabytes of RAM.
MARS' data-translation engines convert data from more than 80 file formats, including popular statistical-analysis packages such as SAS and SPSS, databases such as Oracle and Informix, and spreadsheets such as Microsoft Excel and Lotus.
MARS is currently available on all supported platforms. The product scales easily from the desktop to enterprise-wide data-mart/warehouse versions. The base price for a desktop version is $2,495 and UNIX and Windows NT server site licenses range from $5,000 to $25,000. Discounts are offered on volume purchases and special terms are available for educational institutions.
Founded in 1983, Salford Systems specializes in data-mining and data-analysis products and training and consultation services for a broad range of industries. Salford Systems also offers statistical consulting and custom programming services for the design and analysis of discrete choice experiments. The company has its headquarters in San Diego, Calif., with representation throughout North America, Japan, Europe and Australia.