When evaluating a [[Machine Learning]] [[Classification|classification]] model, **recall** (or sensitivity), on the other hand, is the fraction of relevant instances that were indeed retrieved. In the context of binary classification, it could be understood and the fraction of existing positives that were predicted as such. $ \textrm{Recall} = \frac{\textrm{TP}}{\textrm{TP} + \textrm{FN}} $ It is very tightly coupled with another [[Error Metrics|error metric]], [[Precision|precision]]. Real life models usually require by their very nature a higher precision than recall or the other way around.