Landmark 4 stops to get here · leads to 3

Recall

Also called Sensitivity

The share of actual positives a model correctly identifies, TP / (TP + FN); also called sensitivity or the true positive rate.

Your route here

4 stops · basics first
  1. Machine Learning ✓ understood

    Building systems that learn patterns from data instead of following hand-written rules, getting better at a task as they see more examples.

  2. Supervised Learning ✓ understood

    Learning from examples paired with the correct answer, so a model can predict answers for new inputs it hasn't seen.

  3. Classification ✓ understood

    A supervised learning task where the model assigns each input to one of a fixed set of categories, such as spam or not spam.

  4. Confusion Matrix ✓ understood

    A table showing true positives, true negatives, false positives, and false negatives for classification evaluation.

  5. Recall · you are here ✓ understood

Picture it

PREDICTEDpositivenegativeACTUALpositivenegativeTP42FN6FP8TN44precision = TP / (TP + FP)= 42 / 50 = 0.84recall = TP / (TP + FN)= 42 / 48 = 0.88
Notice recall only looks at the actually-positive row: of all real positives, TP/(TP+FN) were caught, and FN are the misses.

Recall answers one question about a classifier: of all the things that really were positive, how many did it catch? A cancer screen with 90% recall finds 90 of every 100 real cancers and misses 10. The formula is TP / (TP + FN): true positives divided by everything that was actually positive, caught or not.

Recall comes from information retrieval, where it measures how many of the relevant documents a search returned. In medicine the same number is called sensitivity, and elsewhere the true positive rate. They are one quantity under three names.

Reading it off the confusion matrix

The diagram highlights the only row recall cares about: the cases that are actually positive. Within that row, each case either landed as a true positive (caught) or a false negative (missed). Recall is the caught share. Everything in the actually-negative row, including false alarms, is invisible to it. See confusion matrix for the full grid.

That blindness is the key to using it well. A model that labels every single input positive has perfect recall, because it can’t miss anything. It is also useless. Recall on its own can always be bought by flagging more.

Recall versus precision

Precision is the mirror image: of everything the model flagged, how much was really positive. The two pull against each other. Lower the decision threshold and the model flags more, so recall rises and precision usually falls. Raise it and the reverse happens. The Stanford IR textbook gives the classic contrast: a web searcher wants a clean first page (precision), while a paralegal or intelligence analyst wants to find everything and will wade through junk to do it (recall).

Which to favour depends on the cost of each mistake:

  • Favour recall when a miss is expensive: disease screening, fraud detection, safety defects, legal discovery.
  • Favour precision when a false alarm is expensive: spam filters that might bury real mail, or alerts that people learn to ignore.

Summarizing the trade-off

The F1 score is the harmonic mean of precision and recall, which is low if either one is low. A precision–recall curve plots both across every threshold, so you can choose an operating point instead of accepting the default. On imbalanced data, where positives are rare, these are far more informative than accuracy, which a model can score highly on by predicting “negative” every time.

Where it sits

Before this

Confusion Matrix
Recall

Leads to

F1 Score Precision-Recall Curve ROC Curve

Explore nearby

Evaluation Precision Of everything a model flagged as positive, the fraction that really was positive: TP / (TP + FP). It answers "when the model says yes, how often is it right?" Evaluation F1 Score The harmonic mean of precision and recall, providing a single metric that balances both concerns. Evaluation Specificity The proportion of actual negatives correctly identified. Evaluation Precision-Recall Curve A curve showing the tradeoff between precision and recall at different thresholds.

Sources

  1. Christopher D. Manning, Prabhakar Raghavan and Hinrich Schütze, Introduction to Information Retrieval, "Evaluation of unranked retrieval sets" . Cambridge University Press, 2008
  2. scikit-learn, sklearn.metrics.recall_score . Official API documentation