topic Modelling for High Event Rate in SAS Data Science

Modelling for High Event Rate

okla — Tue, 18 Apr 2017 17:26:03 GMT

Hi,

I have a dataset (30k) with event = 82% and non-event = 18%, is this still possible to use normal way to model (ie. neural/regression/etc)? So far, i dont' have a good result yet (DT, regression, scorecard, 50/50 oversampling, Neural was used). Or should i try to reverse the Event to 18% and build a model for that?

Re: Modelling for High Event Rate

Reeza — Tue, 18 Apr 2017 17:28:31 GMT

18%/82% is actually a good split in my experience.

Why do you think you don't have a good fit?

Re: Modelling for High Event Rate

okla — Tue, 18 Apr 2017 17:35:42 GMT

All the models give me a very unstable lift chart with a lift of 1.10 at most. I am still trying out with different segments, but this will only reduce my 30k to a smaller size, which i am trying to avoid.

Re: Modelling for High Event Rate

Ksharp — Wed, 19 Apr 2017 06:24:44 GMT

This is what Data Mining should be if you don't have a big table.
Why not try PROC LOGISTIC ,since your table is quite small .