searching the forum I have not found what I am looking for. Therefore, I hope that anyone here can help me.
I apply the "Threshold Finder" and "Threshold Applier" as post-processing steps with a two-class classification setup. From my point of view, the setup works perfectly.
But I do really need to gain in-depth understanding on how the "Threshold Finder" works. The tutorial simply states that the thresholds are found given "prediction confidences, costs and distributional information". Do you have any idea how these factors actually influence the threshold (with unequal costs for classes positive and negative)? How are these implemented in RM? Any suggested readings?
I think the easiest thing would be to check out rapid miner from sourceforge (see our website for details) and take a look at the thresholdfinder in the class com.rapidminer.operator.postprocessing.ThresholdFinder. Unfortunately I don't know how its implemented in detail.