🎯  Maximum Calibration Error (MCE)

Maximum Calibration Error (MCE)#

The largest gap between confidence and accuracy across calibration bins.

Important

✨ AI-generated content. This page was written with the assistance of an AI language model and is provided as a learning aid. Despite careful review, it may still contain mistakes, omissions, or out-of-date information. Whether you are new to the topic, a team lead, or a senior practitioner, treat it as a starting point rather than an authoritative reference: read it critically and independently verify anything you act on (code, commands, figures, and factual claims) against official documentation and primary sources before relying on it.

What it is#

Maximum Calibration Error reports the worst calibration gap rather than the average — the largest difference between accuracy and confidence over all bins:

\[\text{MCE} = \max_{m}\,\big|\mathrm{acc}(B_m) - \mathrm{conf}(B_m)\big|.\]

Where ECE asks how miscalibrated on average?, MCE asks how bad does it get?

When it matters#

MCE is the right lens for safety-critical systems — medical, autonomous, financial — where a single badly miscalibrated confidence region can cause harm, even if the average looks fine. Lower is better, as with ECE.

Its limits#

Like ECE it is binning-dependent (the answer shifts with bin count and scheme), and it is not a proper scoring rule — a model can achieve low calibration error with trivial predictions, so MCE must be read alongside discrimination metrics, not alone.


Theme: Probability Calibration  ·  All terminology



See also

Source article Adapted (context, re-expressed) in our own words from: Maximum Calibration Error (MCE) (insightful-data-lab.com).

Tags: purpose: reference topic: terminology level: advanced