Precision
Also: positive predictive value
True positives divided by all predicted positives, TP / (TP + FP). High precision means few false alarms; it usually trades off against recall as the decision threshold moves.
Precision = TP / (TP + FP)
Example
A spam filter flags 50 emails and 40 of them are spam: precision = 40 / 50 = 0.8.