Calibration

Calibration helps us understand how honest a model’s probability outputs are. If the model assigns 80% probability to a set of cases and those cases are actually correct about 80% of the time, the model is well calibrated. This matters greatly in risk scoring, medical decision systems, credit modeling, and trust-sensitive AI applications. A model may have strong discriminative ability and still be poorly calibrated. Calibration therefore makes visible the difference between “being correct” and “being correctly confident.”