Recall
Last Updated: July 29, 2026 | By Mihail Sebastian | AI Dictionary
The share of actual positive cases a model correctly identifies, also called sensitivity – the metric to watch when missing a case is the costly error.
What is Recall?
Recall is the share of actual positive cases that a classification model correctly identifies: true positives divided by all the cases that were genuinely positive. It also goes by sensitivity or true positive rate.
A model with high recall misses little. What recall does not measure is how many false alarms the model raised along the way; that side of the ledger belongs to precision.
How Recall Works
Every genuinely positive case ends up either found (a true positive, TP) or missed (a false negative, FN). Recall is the fraction found:
\[ \text{Recall} = \frac{TP}{TP + FN} \]Both counts sit in the confusion matrix. Recall alone is easy to game: a model that flags every single case reaches 1.0 while being worthless. It is always read next to precision, and recall at each decision threshold forms the vertical axis of the ROC curve.
Recall vs Precision
Recall asks how many of the real positives the model found; precision asks how many of the model’s flags were right. Tuning a model to flag more aggressively raises recall and lowers precision; tuning it to flag cautiously does the opposite.
| Recall | Precision | |
|---|---|---|
| Question answered | Of all real positives, how many were found? | Of everything flagged, how much was right? |
| Error it penalizes | False negatives (misses) | False positives (false alarms) |
| Denominator | All actual positives (TP + FN) | All predicted positives (TP + FP) |
| Matters most when | A miss is expensive | A false alarm is expensive |
In cancer screening, a missed tumor is far worse than an extra follow-up test, so recall leads. In spam filtering, losing a real email hurts more than seeing one spam message, so precision leads. When neither error is clearly cheaper, the F1 score averages the two.
Example of Recall
A hospital screens 1,000 patients; 50 actually have the disease. The model flags 80 patients and catches 40 of the 50 real cases, missing 10.
Recall is 40 / 50 = 0.8. Ten sick patients go home with a clean result, and no amount of precision elsewhere repairs that. If the screening exists to catch the disease early, those ten false negatives are the number the hospital cares about most.
Related AI terms: Precision · F1 Score · Confusion Matrix · ROC Curve · Model Evaluation
Did you like the Recall gist?
Learn about 250+ need-to-know artificial intelligence terms in the AI Dictionary.
Mihail Sebastian — Writes about AI governance, regulation, and the technology behind them. Placeholder bio — replace with a real credential line. About