Decision Tree - Enterprise Miner

[jaycette]

I'm making a decision tree in SAS EMiner. I have the data source, then the data partition node, and then a filter node before the decision tree node. When I try a decision tree interactively, some of the splits aren't making sense to me. For example I have one variable that is nominal and can be either "S" or "U". When the tree chooses this variable to split on it has two branches: "S" and "Missing Values Only". None of the entries are missing this variable. I thought that maybe the "U"'s were being filtered out by default because there were too few of them, but when I changed the "Default Filtering Method" to "None" the results did not change. I tried to change the seed number in the data partition node, but that didn't seem to help either. Does anybody know why this might be happening, or how I can fix it?

Thanks,

[PaigeMiller]

My experience is that when SAS says there are missing values and the user says there are no missing values, I believe SAS. So you need to look at your data carefully, or do a PROC FREQ on this particular variable (or whatever the equivalent is in Enterprise Miner).

[jaycette]

It's not that SAS is saying there are some missing values, it is only identifying the "S"'s and the rest as "Missing", so I'm not sure what's happening to the "U"'s. I don't know what what PROC FREQ is, I'm new to SAS and have been working primarily with Enterprise Miner.

[PaigeMiller]

Still, you need to look at your data set somehow and see exactly what is in there.