Software

"Did you ever suspect your additive scorecards to miss important interactions? Or did you dislike your rule-based scorecard for missing the additive element? The BrainMine Scorecard Builder allows for both additive and interactive elements in modeling," says Clemens van Brunschot, CEO of the BrainMine Company (www.brainmine.nl).

How does this work?

The Scorecard Builder is an inexpensive set of SAS macros and a GUI that sits on top of SAS, and is based on automatic optimal binning (a type of transformation) of all types of predictor variables. This transformation is aimed at prediction of a binary target variable like risk, response or purchase. For the actual modeling SAS Proc Logistic is invoked. This logistic regression modeling results in a classical and often powerful scorecard where positive and negative score points are associated with bins (ranges, buckets) of predictor variables. The score is established by adding the score points. This is the additive element of the procedure.

But before the scorecard is produced, the binned variables can be entered into the Interact module. That will perform a search for interactions between predictor variables to arrive at a still better prediction of the target variable. The decision tree algorithm used is the free Treedisc macro written by SAS Institute.

The GUI and the parameter setting give the user control over e.g.:

• width and depth of interaction search;
• selection of cases;
• significance level of splits;
• minimum size of groups before and after splitting;
• criteria for assessment and acceptance of an "interaction";
• naming of resulting interaction variables.

Van Brunschot: "Request your free trial copy of the BrainMine Scorecard Builder at www.brainmine.nl and find out how much money you can save with it."