Precision

Precision measures how often a model is correct when it predicts the positive class. It is calculated as true positives divided by all predicted positives, including false alarms. High precision matters when acting on an incorrect positive is costly, such as blocking a legitimate transaction or wrongly flagging medical evidence. Precision can be increased by raising a classification threshold, but that often lowers recall because more true cases are missed. The value also changes with class prevalence, dataset selection, and the definition of the positive outcome. It should be reported with recall, confusion-matrix counts, thresholds, and subgroup analysis rather than presented as an isolated measure of overall quality.