topic Re: How to reduced efficiently the Class levels in the Association & Decision nodes in SAS Data Science

How to reduced efficiently the Class levels in the Association & Decision nodes

MirzaShaheenBaig — Tue, 25 Nov 2014 17:32:03 GMT

Hi Bees,

I am new to Eminer 6.2 version, would like to use Association & Decision nodes but eminer throwing error, to reduced the Class levels and it is reached to maximum class level 512 exceeded.

Could you please help me out how to handle efficiently class level below 512 level.

Thanks in advance for you valuable advises.

Regards,

Mirza

Re: How to reduced efficiently the Class levels in the Association & Decision nodes

M_Maldonado — Tue, 25 Nov 2014 18:18:14 GMT

I would suggest unsupervised learning like Clustering. That would be my go-to because it is quick, and also easy to explain.

If you want to try more advanced things, Andrew Cathie has presented papers recently at SAS Global Forum on using the outputs of decision trees and regressions to solve the curse of dimensionality. Links to his papers:

The Anti-Curse: Creating an Extension to SAS® Enterprise Miner™ Using PROC ARBORETUM

http://support.sas.com/resources/papers/proceedings11/155-2011.pdf

Extension Node to the Rescue of the Curse of Dimensionality via Weight of Evidence (WOE) Recoding

http://support.sas.com/resources/papers/proceedings13/014-2013.pdf

I hope it helps,

Miguel

Re: How to reduced efficiently the Class levels in the Association & Decision nodes

MirzaShaheenBaig — Tue, 25 Nov 2014 20:42:27 GMT

Thanks Miguel for your advise.

I will explore this reference.

Regards,

Mirza