Confusion matrix in KNIME and WEKA

Dear friends,

I am comparing the confusion matrix obtained using CostSensitive in KNIME with respect to that obtained using WEKA and I have observed that in some case, there are significative differences between them. However, both programs select the same set of features (descriptores) to build the models.

Could anyone explain the reason of these differences?

Thank you a lot in advance for your time and consideration,




Hi Marta, 

Would it be possible for you to upload  a pair of examples that demonstrate this difference (one from  KNIME and the corresponding analysis in WEKA)?

Best Regards