Precision
Of the cases the model predicted as positive, how many actually were?—measuring false alarm rate.
Think of it like
Like the accuracy of a smoke detector: if it goes off, is there actually fire?
Example
A fraud detector flags 100 transactions as fraud. 90 are actually fraud. Precision is 90%.
How it actually works
High precision means few false positives. Low precision means the model cries wolf a lot. Precision alone is incomplete—you also need recall. A model that predicts 'not fraud' for everything has perfect precision (vacuously—it makes no positive predictions) but catches no fraud.
For product teams
For user-facing alerts (fraud, toxicity), prioritize precision to avoid annoying false alarms.
For engineers
Track precision-recall tradeoffs; move the decision threshold to adjust the balance.
Related
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome