Hi!
I'm working on a thesis to analyze the performance of a decision tree classifier (I use the j48 weka for the model).
I use a pattern like that in pictures.
Now, I have to achieve a cascade of decision tree classifiers.Someone knows a scheme in knime to perform this project? Thanks
predictor1 is a simple decision tree predictor ( wich I have modified, but it isn't important)
Hello Antonio,
I'm referring to a 'cascade of classifiers' as mentioned in this paper:
http://research.microsoft.com/~kumarc/pubs/chellapilla_drr06.pdf
There is no meta-node to enable this functionality automatically. However, you could do the following:
place a RowFilter-Node 1 after each prediction node and filter out all examples that have been predicted with confidence > t
place a second RowFilter-Node 2 after the prediction node to obtain all samples with a prediction confidence < t
place a column filter node after RowFilter-Node 2 and filter out the class column and the class probability columns. This should be the input for the next section.
..and so on.
However, you need a classification/prediction node that outputs class probabilities (the DecisionTreePredictor does not). But you can use the weka nodes or the MultiLayerPerceptron/Bayes from KNIME.
Hope this helps..
Nicolas
[/]