I am trying to introduce a model built with WEKA3.6 into KNIME. The model was built using Z-normalized 2D MOE descriptors, feature selection and Costsensitiv in combination with Random Forest. These 3 last features have been included in the Attribute Selected Classifier node.
Using KNIME, I reproduce the same descriptor values for the Z-normalization and I obtain the same set of descriptors as those selected using WEKA directly. However, the confusion matrix is different. I have tried many options (use the arff files obtained directly from Weka, add the activity column at the last position, etc), but nothing works.
Does anyone know what is happening? I woul be very thankful for any help or suggestion.
Thanks in advance for your time and consideration.