How to calibrate the threshold in workflow about Feature Elimination?

Hello everyone.

I downloaded the "002004_Features-Elimination-with-Naive Bayes" example, located in the 002_Datamining folder. If I want to automatically select the attributes in a predictive model, what value should I give the threshold and what is the reason for that value? If I put a value close to 0.3 do not choose me any attribute, if under the value to 0.1, choose 10 attributes, but do not know why.

Does anyone KNIME friends can enlighten me the way?

Thank you very much.