07-26-2016 06:52 PM
Unfortunately I am not a SAS Programmer, but use the Enterprise Miner diagrams along with the nodes. I haven't been able to find any good demonstrations of the survival analysis nodes in use.
I have a list of 20K folks that have retired, and 30k folks that are eligble to retire and want to predict when these eligibles will retire... I have calculated in months how long those that retired worked after they were eligible to retire. Althogh a good few leave on the first day eligible, we have thousands that continue working years and years after they are eligible to retire. I am using months as the target, and in running a typical model, the best performing Auto Neural network doesn't give me good predictions on when each of these 30k will leave
Anyone any examples/idea on how I can apply this through survival analysis? Or is there a better method?
07-27-2016 12:07 AM
Survival analysis through EM is probably overkill, Base SAS does a good job.
You our can try Proc Phreg and LIFEREG. You can also add a code node and try these in EM as well. It doesn't quite sound like survival data though IMO, can't see where the censoring is applicable. Censoring may not be required.
07-27-2016 08:58 AM - edited 07-27-2016 08:58 AM
This tip might help you out:
It includes links to our GitHub repository that has example EM diagrams (templates) using the Survival node and sample data that you can run to get an idea of how it works, as well as videos describing the survival data mining approach. Since your goal is prediction, I think this might be a better approach than PHREG or LIFEREG.