Is there a way to establish a minimum threshold (of accuracy) for accepting a rule in a Decision Tree?
Ideally I would have a decision tree that would classify all outcomes in 0, 1, or Indefinite. The Indefinite set would be populated by the cases in which the rule(s) did not reach a minimum % of accuracy.
By default the decision tree predictor simply picks the prediction with the highest probability, but you can change that however you like. In its dialog, enable “Append columns with normalized class distribution”. This will give you the probability for each class in multiple appended columns. Now you can use a Rule Engine node to derive your own prediction from that.
This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.