Rice Analytics

Automated Reduced Error Predictive Analytics

Risk Management
The quality of your predictive models can be one of your greater risks in risk management applications such as credit scoring or insurance or fraud applications.  This is especially true because the inherent arbitrary nature of variable selection in Stepwise Logistic Regression means that each modeler will be likely to generate a different predictive model due to arbitrary parameter cutoff choices and because Stepwise regression often returns regression coefficients with incorrect signs.  Any attempt to fix these incorrect signs may be very risky, but leaving these signs in the model is also risky.  Reduced Error Logistic Regression (RELR) removes this source of risk and uncertainty from your risk management program, as all modelers will generate the very same parsimonious PARSED RELR variable selection with correct signs.  Furthermore, RELR models usually will be much more accurate than Stepwise Logistic Regression models when you have many important variables and interactions, correlated variables, nonlinear variables, and/or small sample sizes.  Please read our Case Studies page for specific case examples of how RELR has been quite beneficial in risk management applications in credit scoring.

The quality of your predictive modeling solution is probably one of the great risks in your risk management program.  Our Reduced Error Logistic Regression will help you to build much higher quality predictive models and lock out much of your analytical risk such as is related to regression coefficients with incorrect signs.

Machine Learning  Segmentation  Consumer Surveys  Predictive Modeling  Risk Management