Precision measures how often predicted positive cases are actually positive. In plain language: when the system raises an alarm, how often is the alarm right? section 8.9
Wolf alarms
A wolf-detection system with 94.7% precision is correct when identifying a wolf 94.7% of the time; the remaining 5.3% of alarms are false. A system with 0% precision makes every alarm a false alarm. section 8.9
Use precision carefully when judging classification systems:
- Precision focuses on predicted positive cases and does not use true negatives. section 8.9
- Compare it with recall: a system can make accurate positive predictions while still missing many real positives. section 8.9
- Check class imbalance before trusting accuracy; a rare event can produce high accuracy even when the system never predicts it. section 8.9
