KDnuggets : News : 2001 : n24 : item12    (previous | next)

Software


From: BrainMine
Date: Sat, 17 Nov 2001 06:21:08 +0100
Subject: BrainMine introduces the Scorecard Builder

"First we developed powerful SAS macros for predictive modeling of risk, response or purchase. Now we added a Graphical User Interface that generates the SAS code to invoke these macros," says Clemens van Brunschot, CEO of the BrainMine Company. "Request your free trial copy of the 'Scorecard Builder' at www.brainmine.nl and find out how much money you can save with it".

The most important features of this software are:

  • Transformation of predictor variables into optimal bins (handling problems like outliers, missing values and non-linear relationships).
  • Hybrid approach: (logistic) regression combined with decision trees to detect interactions, all in the same scorecard.
  • Automatic translation of the resulting model into SAS data step code in terms of the original predictors.
  • Therefore: easy application of the model in a test or evaluation dataset.
  • Using the power of SAS, e.g. for documentation and replication of the modeling process.
  • Automation of all the tedious steps in modeling, but:
  • Many points in the modeling process where the analyst can add experience and knowledge to guide the process.
There used to be a time when early versions of the underlying SAS macros were available in the public domain. Since then many improvements have found their way to the software as an answer to the needs of users, e.g.:
  • Flexibility in the definition of 'interaction'.
  • Setting up multiple decision trees to detect more interactions.
  • Model translated to the original predictors (including interactions).
  • Improved bin management: minimum bin size, generalization of upper and lower bin limits respecting SAS/other missings.
  • Increased speed of the binning process.
  • A macro to reduce multicollinearity among the predictors (built together with Fingerhut.com).
  • GUI.

KDnuggets : News : 2001 : n24 : item12    (previous | next)

Copyright © 2001 KDnuggets.   Subscribe to KDnuggets News!