Precision and recall
Beginner
Two ways to score an alarm. Recall: how many real problems it caught. Precision: how many of its alarms were real.
Novice
For a model that flags problems: recall is the share of real problems it flagged, and precision is the share of its flags that were real problems. Flagging more raises recall and usually lowers precision.
Expert
Recall = TP / (TP + FN), precision = TP / (TP + FP). With rare positives, a model that never flags anything has high accuracy and zero recall, so papers report recall at a fixed false-positive rate, F1, or a whole precision–recall or ROC curve.
See also: DRC hotspot.