Webcasts

Date and Time of Webinar: Friday April 17, 9 am PDT, 11:00 AM CDT, noon EDT
Duration: 1 hour
Presented by Daniel M. Rice, Ph.D.

Reduced Error Logistic Regression (RELR) is a new and general form of regression that avoids problems related to error and dimensionality. RELR completely automates tasks related to feature extraction, feature selection, model balancing, and error reduction, so RELR models are extremely easy to build. With difficult error-riddled problems involving a large number of multicollinear variables that typically require a relatively large sample size, RELR can perform significantly better than state-of-the-art predictive analytics methods such as Support Vector Machines (SVM). With easier problems, RELR appears to perform comparably to state-of-the-art methods such as SVM. However, unlike SVM and other such nonparametric methods, RELR is completely transparent and parametric. This allows parsimonious and interpretable RELR models where the estimated odds of an event can be easily visualized as a function of each variable's values across all observations.

This webinar is a free tutorial that will be presented by Daniel M. Rice, Ph.D. Part of this same webinar has been presented at several major conferences over the past few years such as JSM, M2007, and MSWSUG. RELR is implemented as a stored compiled SAS macro that can run within SASŪ or snap in as an Enterprise Miner™ extension node. Yet, running a RELR model simply requires one to fill out a list of model design parameters; this can be done without any knowledge of SAS programming. Therefore, this webinar should be of general interest to those in the predictive analytics community whether they are users or managers and whether or not they have SAS experience.

The webinar will consist of the following components:

• An overview of how error is modeled and reduced in RELR
• An overview of how RELR handles the 'curse of dimensionality'
• An overview of RELR's Parsed™ variable selection mechanism and the strengths and weaknesses of Parsed RELR models vs. Full RELR models
• Specific modeling results where RELR has performed better than other predictive methods vs. roughly the same as other predictive methods in terms of fit statistics
• A demonstration of the input parameter screens used to run a RELR model

SASŪ and Enterprise Miner™ are trademarks of SAS Institute. Parsed™ is a trademark of Rice Analytics. Rice Analytics is a member of the SAS Alliance Partnership.