PROC SMOTESAMPLE

[phopkinson]

Hi,

I'm currently using the smotesample action in SAS and encountering two unusual issues:

1. Unexpected Categorical Values:
   Some of the newly generated categorical variables—are returning large negative values (e.g., -1000000) that do not exist in the original dataset. Could you clarify why this might happen? Specifically, how does SMOTE handle categorical features, and is there a recommended approach to ensure realistic synthetic values?

2. Amplification of Placeholder Values:
   My dataset uses placeholder values like -1 or -2 to represent missing data. After applying SMOTE, these placeholders seem to become more dominant in the sampled data. Is there a way to prevent SMOTE from oversampling these values or to exclude them from the interpolation process?

Any guidance or best practices would be greatly appreciated.

[Kathryn_SAS]

Please include your SAS log that shows the code you are running and any messages you are getting. If you can provide sample data that would also be helpful.