Minimum-Risk Recalibration of Classifiers

A probabilistic classifier is considered "well calibrated" when its predicted probabilities closely align with the empirical frequencies of the corresponding labels [