Step by Step
P
The formula — TP / (TP + FP)
Precision asks: out of everything the model predicted as positive, what fraction was actually correct? Formula: True Positives divided by (True Positives plus False Positives).
Example: if a spam filter flags 100 emails as spam and 90 of those are actually spam, precision = 90/100 = 90%.
R
The sniper metaphor — only fires when sure
Like a sniper who only takes a shot when highly confident of hitting the target, high precision means the model rarely makes a positive prediction unless it's very likely correct — minimizing false alarms.
Example: a high-precision spam filter almost never flags a legitimate email as spam, because it only "pulls the trigger" when quite confident.
⚠
When precision matters most — costly false positives
Precision should be prioritized specifically when false positives carry a high cost, such as flagging innocent people as criminals, or blocking legitimate, important emails as spam.
Example: a criminal justice risk-assessment tool should prioritize precision, since falsely flagging an innocent person carries severe real-world consequences.
Applied Walkthrough
1
A spam filter flags 100 emails as spam over the course of a week, and upon manual review, 90 of those are confirmed to actually be spam.
2
Ask: what is this spam filter's precision? 90/100 = 90%, since precision measures TP divided by all predicted positives (TP + FP).
3
Ask: does this tell us anything about how many actual spam emails the filter MISSED? No — precision only measures the reliability of what it DID flag, not how comprehensive it was.
4
This is exactly why the sniper metaphor fits: precision measures whether the shots taken (positive predictions) hit their target, not how many targets were engaged in the first place.
Exam Application
Exams test whether you can correctly compute precision from a given confusion matrix and whether you can identify scenarios where high precision matters most — specifically when false positives carry a high real-world cost.
⚠ Common Trap
The most common trap is confusing precision with overall accuracy, or assuming high precision alone tells you the model catches most real positives. Precision says nothing about how many actual positives were missed (that's recall's job) — a model could have perfect precision while still missing the vast majority of real positive cases.
✓ Quick Self-Check
1. What formula calculates precision?
TP divided by (TP + FP) — true positives over all predicted positives.
Tap to reveal / hide
2. What does the sniper metaphor represent about precision?
Only firing (predicting positive) when highly confident, minimizing false alarms — like a sniper who only shoots when sure of a hit.
Tap to reveal / hide
3. Does high precision guarantee the model catches most actual positive cases?
No — precision only measures the reliability of the positive predictions made, not how comprehensive the model was at catching all actual positives.
Tap to reveal / hide
4. When should precision be prioritized?
When false positives are especially costly, such as falsely accusing an innocent person or blocking a legitimate email.
Tap to reveal / hide
5. If a filter flags 50 items as positive and 45 are correct, what is its precision?
45/50 = 90%.
Tap to reveal / hide