In information retrieval and classification, the fraction of items labeled as positive by a system that are actually true positives; also called positive predictive value.

Principle

Quantify the trustworthiness of positive predictions by dividing true positives by all predicted positives; emphasizes correctness among predicted positives rather than coverage of all actual positives.

Demonstration

A spam filter marks 50 emails as spam; 40 of those are truly spam, so precision = 40/50 = 0.8.

Misapplication

Optimizing solely for precision can produce a system that labels very few items as positive (to avoid false positives), yielding poor recall and missing many actual positives.

Consequence

High precision reduces the rate of false alarms and is appropriate when the cost of false positives is high; it supports decisions where positive predictions must be reliable.

Reversal

Prioritizing recall over precision (e.g., by lowering the decision threshold) increases detection of positives but typically reduces precision because more false positives are included.

Boundary

Defined for classification tasks with a clear positive class and binary or thresholded outputs; not directly applicable to multi-class settings without per-class definition or to regression without discretization.

Semantic Tension

Often contrasted with accuracy and recall; precision measures quality of positives while recall measures quantity of true positives captured—balancing both is necessary for reliable classifiers.

Synthesis

Precision measures the purity of predicted positives: the proportion of predicted positives that are correct. Use it when false positives are costly, and combine with recall or F-score to understand overall performance.