Glossary

Precision and recall

Two complementary classifier metrics; precision is how many predicted positives were correct, recall is how many actual positives were found.

Also called: sensitivity, positive predictive value

Precision and recall are two measures of a classification model's accuracy on the positive class, read directly from a confusion matrix. Precision is true positives / (true positives + false positives): of everything the model flagged as positive, how much actually was. Recall, also called sensitivity, is true positives / (true positives + false negatives): of everything that actually was positive, how much the model found.

The two trade off against each other for a given model: making a fraud detector flag more transactions as suspicious tends to raise recall, catching more real fraud, while lowering precision, generating more false alarms. Moving the classification threshold shifts that balance in one direction or the other. Plain accuracy hides this trade-off entirely and is a poor metric on its own, especially when the positive class is rare.

Which measure matters more depends on the cost of each error type: a cancer screening test favors high recall, since missing a real case is costly, while a spam filter favors high precision, since blocking a legitimate email is disruptive. The f1 score combines both into a single number when neither error type clearly dominates, and ROC AUC summarizes performance across all possible thresholds at once.

Last reviewed September 22, 2026

In the index now

Related terms

Confusion matrix A table comparing a classifier's predicted labels against actual labels, breaking results into true and false positives and negatives. F1 score The harmonic mean of precision and recall, giving a single score that balances both when neither error type is clearly more costly. ROC AUC (area under the ROC curve) A single number summarizing how well a classifier separates positive from negative cases across every possible decision threshold. Classification A supervised learning task that assigns each input to one of a fixed set of discrete categories or labels. False discovery rate (FDR) The expected share of "significant" results that are actually false positives, when many statistical tests are run and checked at once.