I generated a dataset and I applied some machine learning tools for classification and regression like random forest and SVM.
I am getting very weird confusion matrix for classification and R2 for the regression. I noticed that that all the test set falls in one or two classes only even when I have 5 or 7 classes.
When I tried the regression, I got negative R2.
I tried to delete some classes with their corresponding samples from the dataset but I still the classes in the confusion matrix. This was not solved even when I put a fresh node for the learner and predictors and scorer (I am using weka nodes).
The only way that worked for me to erase the classes from the scorer was to save the dataset to a csv file and call it afain in a reader node.
Is there a reason for this weird behaviour in Knime (it was working well until recently)
After you filtered out the classes you can use the Domain Calculator node to update the table specification. This way only the still available values in the table are valid and possible values for your prediction model.
Just have a look at the table spec before and after applying the filtering and domain calculation. You can check the table spec on the second tab of the output data view.