Step by Step
R
The formula — TP / (TP + FN)
Recall (also called sensitivity) asks: out of everything that was actually positive, what fraction did the model correctly catch? Formula: True Positives divided by (True Positives plus False Negatives).
Example: if 100 patients actually have cancer and the model correctly identifies 90 of them, recall = 90/100 = 90%.
N
The fishing net metaphor — trying to catch everything
Like casting a wide fishing net to catch as many fish as possible (even if some unwanted bycatch gets swept up too), high recall means the model sweeps wide to catch nearly every real positive case, even at the cost of some false alarms.
Example: a high-recall cancer screening test catches nearly every actual cancer case, even if it means flagging some healthy patients for unnecessary further testing too.
⚠
When recall matters most — costly false negatives
Recall should be prioritized specifically when false negatives (missed detections) carry a severe cost, such as missing an actual cancer diagnosis or missing an actual fraudulent transaction.
Example: a cancer screening test should prioritize recall, since missing an actual cancer case (a false negative) is far more dangerous than a false alarm requiring extra testing.
Applied Walkthrough
1
Among 100 patients who actually have a particular disease, a screening model correctly identifies 90 of them, missing the remaining 10.
2
Ask: what is this model's recall? 90/100 = 90%, since recall measures TP divided by all actual positives (TP + FN).
3
Ask: does this tell us how many healthy patients were incorrectly flagged as sick? No — recall only measures how comprehensively actual positives were caught, not how many false alarms occurred among the negatives.
4
This is exactly why the fishing net metaphor fits: recall measures how much of the actual "school of fish" (real positives) got caught in the net, regardless of how much unwanted bycatch (false positives) also got swept up.
Exam Application
Exams test whether you can correctly compute recall from a given confusion matrix and whether you can identify scenarios where high recall matters most — specifically when false negatives (missed detections) carry a severe real-world cost.
⚠ Common Trap
The most common trap is confusing recall with precision, or assuming high recall alone tells you the model's positive predictions are trustworthy. Recall says nothing about how many false alarms occurred among the predicted positives (that's precision's job) — a model could have perfect recall while still generating many false positives.
✓ Quick Self-Check
1. What formula calculates recall?
TP divided by (TP + FN) — true positives over all actual positives.
Tap to reveal / hide
2. What does the fishing net metaphor represent about recall?
Sweeping wide to catch nearly every actual positive case, even at the cost of some false alarms — like a net catching bycatch along with the target fish.
Tap to reveal / hide
3. Does high recall guarantee the model's positive predictions are highly trustworthy?
No — recall only measures how comprehensively actual positives were caught, not how many false alarms occurred among the predicted positives.
Tap to reveal / hide
4. When should recall be prioritized?
When false negatives (missed detections) are especially costly, such as missing an actual cancer diagnosis or fraudulent transaction.
Tap to reveal / hide
5. If 50 actual positive cases exist and a model correctly identifies 40 of them, what is its recall?
40/50 = 80%.
Tap to reveal / hide