Beyond sigmoids: how to obtain well-calibrated probabilities from binary classifiers with beta calibration (Q1688980)

A predetective model is said to be well-calibrated if its predictions match observed distributions in the data. Often used method of the calibration is a logistic calibration. A new type of calibration like calibration based on the beta distribution is considered. It has been shown that implementation of the beta calibration is easy. Several experiments are used to prove the raison d'etre of the new method.