Precision is an evaluation metric that measures how many of a model's positive predictions are actually correct. It answers a focused question: when the model says yes, how often is it right? High precision means the model produces few false positives.
Precision measures the quality of a model's positive predictions. Of all the times a model predicted the positive class, precision is the fraction that were actually correct. It is defined as true positives divided by the sum of true positives and false positives.
Precision says nothing about the positives a model missed, only about how trustworthy its positive calls are. That makes it one half of a pair with recall, which measures how many of the real positives the model actually found.
Key takeaways
Precision answers "when the model predicts positive, how often is it right?"
It is calculated as true positives divided by all predicted positives.
High precision means few false positives, but it says nothing about missed detections.
What precision measures
The components that define precision.
The components that define precision.
Component
What it means
True positives (TP)
Positive predictions that are correct
False positives (FP)
Positive predictions that are wrong
Precision
TP divided by (TP plus FP)
How it works
You take every prediction the model labeled positive, then check each against ground truth. Precision is the count of correct positives divided by the total number of positive predictions. In detection tasks, a prediction usually counts as correct only when it matches a ground-truth object above an intersection over union threshold, so the choice of threshold affects the score. Lowering the model's confidence threshold typically surfaces more positives and lowers precision, while raising it does the reverse.
Why it matters
Precision is the metric to watch whenever a false positive is expensive. A defect detector that constantly cries wolf wastes human review time, and a screening model that over-flags erodes trust. Because precision trades off against recall, most teams tune a decision threshold to balance the two for their use case, or track both together through the precision-recall curve rather than optimizing either in isolation.
Frequently asked questions
What is the difference between precision and recall?
Precision measures how many predicted positives are correct. Recall measures how many actual positives were found. Precision punishes false positives, recall punishes false negatives.
When should you optimize for precision?
When false positives carry a high cost, such as flagging content for removal or triggering an expensive downstream action on every positive prediction.
What is a good precision score?
It depends entirely on the task and the cost of errors. Precision is most meaningful when reported alongside recall, since either one is easy to maximize at the other's expense.