I think you may have something w.r.t the problem I am looking to solve, so thought of giving a buzz.
I am in search for some good techniques (or combination of those) to model rare events (a ratio of 1:500 to 1:1000). Apart from logistic (and versions) and boot strapping, I am looking to:
1. Use a combine of logistic and decision tree (say decision tree segmentation followed by logistic on part of data where good-bad ratio is more balanced).
2. Explore neural net
3. Other procedures in SAS.