10-30-2013 08:10 AM
I want to build a response model but with a very low response rate, around 0.2%. I haven't built any model with such low response in the past, so I don't know what's the best way to proceed in this case with SAS Enterprise Miner. I understand the idea of oversampling but as I haven't applied it in the past, your help would be much appreciated. Once you have built the model using oversampling, how do you undo the oversampling? And at the time of scoring new data, which probabilities do we use? The one build on oversampled data or the original??
Is the Oversampling technique recommended? When is it compulsory to use it?
10-30-2013 11:12 AM
I came across a decent blog post about oversampling. Perhaps it contains some answers you seek. http://www.data-mining-blog.com/tips-and-tutorials/overrepresentation-oversampling/
Also check out any links within the post.
10-30-2013 12:48 PM
Here is a link that has video that addresses oversampling with EM
Another SAS note provides step-by-step instruction
Enjoy. Post back if you have questions.