Is there a way to optimize a model (classifier) to a certain user defined metric? The only shuch node I found was the ThresholdSelector from WEKA 3.7 but it does not offer the desired metric.
Have you seen the optimization extension that is part of KNIME Labs? With this extension you can construt a special loop that does precisely as you describe.
Thanks. Theoretically that works but the Weka nodes do not allow flowVariables for setting the parameters. I can only set the class and "serilaized-weka classifier". What is the later? This? http://weka.wikispaces.com/Serialization
Because a scheme string in the form of ".RandomForest -I 30" does not work.
I don't have a great solution then, sorry.
I would also point you to the Decision Tree Ensemble nodes in KNIME Labs. They work nicely with the optimization loop and offer similar functionality to the Random Forest model in WEKA.
Keep us posted on your progress.
I'm making my first steps with maschine learning and for that I went to kaggle and experimented with the https://www.kaggle.com/c/higgs-boson competition. Sadly is does not seem to be possible to acieve good results in KNIME. I went with plain WEKA in Java.