Inference, estimation, and decision-making from data
A classifier often outputs more than a label. Ask it for the chance an email is spam and it might answer 0.8. That number should mean something you can test: among many emails the model rates near 0.8, roughly 80% of them should actually be spam. When predicted probability and observed frequency agree like that, the model is calibrated.
🔒 This is a Pro lesson — the interactive figure, worked examples, quiz and practice open with Pro access.
▶ Calibration & Proper Scoring