Consolidation of categorical inputs

[pvareschi]

Re: "Applied Analytics Using SAS Enterprise Miner", "Lesson 5: Regression Models Using SAS Enterprise Miner"

Is there a rule of thumb in terms of how many levels should be considered too many? Would it be reasonable to say that if using a Regression node, it would make sense to have no input taking up more than 8-10 degrees of freedom?

Or, for predictive modelling, is it ok to allow more levels for categorical variables?

[gcjfernandez · Accepted solution]

I hate to provide rule of thump when optimal solutions for most options are data specific.

However, for this case SAS Enterprise miner advanced metadata advisor is using 20 as the categorical levels threshold to reject the nominal variable. You could consider this default setting as the best rule of thump.