Precision
In one sentence Precision is the fraction of your model's positive calls that were actually correct.
Updated
Precision answers one question: of everything the model flagged as positive, how much really was positive?
Think of a security guard at a wedding who stops people he suspects are gatecrashers. If he stops ten people and eight really were gatecrashers, his precision is 80%. The two innocent uncles he embarrassed are the cost of low precision: false alarms.
The formula is short. Precision = true positives ÷ (true positives + false positives). A true positive is a correct positive call. A false positive is a positive call that turned out wrong.
When precision is the metric that matters
Care about precision when a false alarm is expensive. A spam filter with low precision sends real job offers to the spam folder. A fraud system with low precision blocks honest customers' cards. In both cases, saying "yes" wrongly hurts more than missing a case.
Precision says nothing about what the model missed. A guard who stops one person, correctly, has 100% precision — while twenty gatecrashers walk past. That blind spot is measured by recall, and the two are usually in tension.
Where to go next
- Full lesson: Model evaluation
- Related terms: recall, f1-score, accuracy, confusion-matrix