Continuous WoE definition error #370
Description
In Tutorial: optimal binning with continuous target there is an error the definition of Weight of Evidence. The formula contains absolute value:
which should be
In the code itself it is implemented correctly, but negative values of WoE in binning_table might surprise someone.
On that note I wanted to ask about Information Value for continuous target. While the definition with absolute value seems sensible (with correction to WoE formula it would have to be adjusted as well) in the source you are citing - FICO Building Powerful, Predictive Scorecards 2014 there is no absolute value. Can you comment on that?
Your paper, OptBinning tutorial and FICO paper (which has to be accessed through wayback machine) seem to be the only places where such generalization is mentioned. Which is a shame because it is quite useful.