recall

Recall (Sensitivity, True Positive Rate)

Of all the actual positive cases, the fraction that the model correctly identified — the model's ability to find what it's looking for.

When you'd see it: Any classification task where false negatives are costly: cancer screening (missing a cancer is worse than a false follow-up), fraud detection (missed fraud becomes a realized loss), safety systems (missing a dangerous condition is unacceptable). Also called sensitivity or true positive rate.

Why it matters: Recall is the right primary metric when false negatives are more expensive than false positives. A cancer screening model needs high recall — better to investigate a false alarm than to miss a real case. Lowering the classification threshold increases recall but reduces precision.

Common mistakes: Maximizing recall without regard for precision. A model that predicts positive for every observation has 100% recall but produces nothing but false alarms. The precision-recall tradeoff must be evaluated together, usually via F1 or the precision-recall curve.

Related precision, F1 score, confusion matrix, AUC