Step by Step
P
Precision — PICKY, only says yes when really sure
Precision measures TP divided by (TP + FP) — out of everything the model flagged as positive, what fraction was actually correct? A "picky" model minimizes false alarms.
Example: a spam filter with high precision rarely flags legitimate emails as spam — when it does say "spam," it's almost always right.
R
Recall — REMEMBERS everything, sweeps wide to catch every positive
Recall measures TP divided by (TP + FN) — out of everything that was actually positive, what fraction did the model correctly catch? A model with high recall minimizes missed detections.
Example: a cancer screening test with high recall catches nearly every actual cancer case, even if it means flagging some healthy patients for further review too.
⚖
The tradeoff — threshold controls the balance
Raising the classification threshold (requiring more confidence before predicting positive) tends to raise precision but lower recall; lowering the threshold does the opposite.
Example: requiring 90% confidence before flagging spam raises precision (fewer false alarms) but lowers recall (more spam slips through undetected).
Applied Walkthrough
1
A cancer screening test needs to decide whether to prioritize precision or recall.
2
Ask: is it worse to miss an actual cancer case, or to flag a healthy patient for unnecessary further testing? Missing an actual cancer case is far worse.
3
This means the screening test should prioritize Recall — REMEMBER to catch every real case, even at the cost of some false alarms.
4
Contrast: a spam filter should instead prioritize Precision — being PICKY to avoid blocking real, important emails, even if that means occasionally letting some spam through.
Exam Application
Exams test whether you can correctly identify which metric (precision or recall) matters more for a given real-world scenario, based on whether false positives or false negatives carry a higher cost. Cancer screening (prioritize recall) and spam filtering (prioritize precision) are classic, frequently tested examples.
⚠ Common Trap
The most common trap is assuming higher precision and higher recall can always both be maximized simultaneously by simply making the model "better." In reality, adjusting the classification threshold creates a direct tradeoff between the two — raising one typically lowers the other.
✓ Quick Self-Check
1. What formula calculates precision?
TP divided by (TP + FP).
Tap to reveal / hide
2. What formula calculates recall?
TP divided by (TP + FN).
Tap to reveal / hide
3. What happens to precision and recall when you raise the classification threshold?
Precision tends to rise while recall tends to fall.
Tap to reveal / hide
4. Should a cancer screening test prioritize precision or recall, and why?
Recall — missing an actual cancer case (false negative) is far more costly than flagging a healthy patient for extra testing (false positive).
Tap to reveal / hide
5. Should a spam filter prioritize precision or recall, and why?
Precision — blocking a real, important email (false positive) is worse than letting some spam through (false negative).
Tap to reveal / hide