Software

By BrainMine's CEO Clemens van Brunschot

Since interest in data mining newcomer BrainMine is rapidly growing, it seems to be a good idea to give an overview of the 2002 plans of this company that promotes inexpensive predictive modeling add-ons for SAS 8 or higher. For more information mail to plans@brainmine.nl or have a (general) look at the website: www.brainmine.nl.

Product development for the BrainMine Scorecard Builder

1. Added will be a module for Reject Inference, the most 'mysterious' element in scorecard construction for credit scoring. Building on the other modules, this new module will estimate the target variable for cases that have none measured. This is particularly helpful when you like your credit scorecard to also deal with applicants who were not accepted before; or a response (or purchase) model to also deal with prospects who didn't receive an offer yet.

2. The model evaluation module will be augmented with a table that shows the score distribution in such a way that decisions on cut-off scores can easily be reached, and also with an enhanced graph of the score
distributions as separated by the target variable (e.g. 'divergence' between
goods and bads in credit scoring).

3. The UNIX version of the software will be introduced.

4. Negotiations are taking place with a USA based partner which could result in adding a variable search tool whose purpose is to search for special model covariates or 'enhancement' variables in large databases. Such variables have the potential to increase the predictive ability of variables in one's final model and allow users to improve their segmented response rates or 'lifts' substantially compared to their current yields or to yields obtained with typical modeling methodologies and current data mining software packages.

- Pricing
Because somewhere in 2002 BrainMine will offer a more comprehensive solution including reject inference functionality and enhanced strategy tools, pricing will - for new clients - be adjusted to reflect the product enhancements. The product will still be considered 'inexpensive' compared to any other solution.

- Consultation and training
BrainMine is pulling together a list of consultancy companies around the world that have the capabilities and drive to assist or train customers in building their own predictive models (and in making a lot of extra profit in doing so) by using the Scorecard Builder and SAS. If they are interested in leads, they are invited to send a mail to consultancy@brainmine.nl.

Added topic: Missing value handling
In the BrainMine Scorecard Builder the user specifies per set of predictor variables a list of what should be regarded as missing values. These missing values are - by a chi square test - statistically assigned to the 'bin' or sub-range where they fit best (or they may constitute a separate bin). After predictive modeling, the scorecard with all its bins and interactions is generated in simple SAS code in terms of the original 'unbinned' variables, where all declared missing values are 100% respected.