seniorSupervised Learning
What is probability calibration and why is it critical in decision systems?
Updated May 17, 2026
Short answer
Probability calibration ensures predicted probabilities match real-world outcome frequencies.
Deep explanation
Many ML models output uncalibrated probabilities. Calibration aligns predicted probabilities with observed frequencies. Methods include Platt Scaling (sigmoid-based) and Isotonic Regression (non-parametric). Calibration is crucial in risk-sensitive domains like healthcare and finance where probability interpretation matters.
Real-world example
Insurance risk models estimating probability of claim occurrence.
Common mistakes
- Confusing classification accuracy with probability quality.
Follow-up questions
- What is a calibration curve?
- Which models are naturally well-calibrated?