Underconfident#
When predicted probabilities are less extreme than the true accuracy warrants.
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 Despite careful review, it may still contain mistakes, omissions, or out-of-date information.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#
A model is underconfident when its predicted probabilities are too low for its actual accuracy — it hedges, claiming less certainty than it deserves. A model that is only 80% confident but 90% accurate is underconfident.
How to spot it#
On a reliability diagram, underconfident points fall above the diagonal (accuracy > confidence), and predictions cluster near 0.5 rather than committing. It is the mirror image of overconfidence, and a single model can be overconfident in some ranges and underconfident in others.
Why it matters#
Though it feels “safe,” underconfidence wastes the model’s discriminative signal — useful, correct predictions get muted probabilities, so thresholds and risk-based decisions under-trigger. Like overconfidence, it is fixed by recalibration.
Theme: Probability Calibration · All terminology
Hint
Mind map — connected ideas
Overconfident · Confidence Level · Classification Probability · Risk-Based Decisions · Binary Cross-Entropy (BCE) · Sigmoid Function
Hint
More in Probability Calibration
Adaptive ECE (Expected Calibration Error with Adaptive Binning) · Brier Score · Calibration quality (Model Calibration) · Expected Calibration Error (ECE) · Isotonic Regression · Maximum Calibration Error (MCE) · Murphy’s Decomposition · Overconfident · Platt Scaling · Reliability Curves (also called Calibration Curves) · Temperature Scaling
See also
Source article Adapted (context, re-expressed) in our own words from: Underconfident (insightful-data-lab.com).