[jhokanson]

But isn't it possible to design multiple models, where the judgement of the "best model" is dependent on your goals (e.g. reduction of false positives vs false negatives)?