seniorSupervised Learning
What is model calibration in supervised learning?
Updated May 17, 2026
Short answer
Model calibration ensures predicted probabilities reflect true likelihoods.
Deep explanation
Some classifiers output probabilities that are not well-calibrated. Calibration aligns predicted probabilities with actual outcomes using methods like Platt Scaling or Isotonic Regression. This is crucial in risk-sensitive applications where probability accuracy matters.
Real-world example
In healthcare, a model predicting 80% disease probability should reflect real 80% occurrence.
Common mistakes
- Assuming all probability outputs are inherently reliable.
Follow-up questions
- What is Platt scaling?
- How do you evaluate calibration?