Classification requires a threshold; however, methods like C-statistic and AUC obfuscate this. Luckily, there is a sensible strategy for imbalanced data thresholding.