Recall

Recall (also called sensitivity or the true positive rate) is a classification metric that measures what proportion of the actual positive cases was correctly identified by the model. Its formula is recall = TP / (TP + FN), where TP are the true positives (hits in the positive class) and FN the false negatives (positives the model let slip).

Where it fits: the confusion matrix

Recall is understood within the confusion matrix, which crosses what is predicted with what is real in four cells: true positives (TP), false positives (FP), true negatives (TN) and false negatives (FN). Recall focuses on the row of the actual positives and asks: of all those that were truly positive, how many did the model catch?

Recall versus precision

Recall should not be confused with precision, computed as TP / (TP + FP), which measures what proportion of what the model flagged as positive really was. The difference is key: recall penalizes false negatives; precision, false positives. And there is usually a tension between them: lowering the threshold to catch more positives raises recall but lowers precision, and vice versa. The F1 score, the harmonic mean of the two, sums up that balance in a single number.

When to prioritize recall

A high recall matters when a false negative is very costly. In medical screening for a serious disease, a recall close to the whole is sought: a false alarm, which a second test will rule out, is preferable to missing a sick patient. The same in fraud detection: letting a fraudulent transaction slip usually costs more than over-reviewing a legitimate one.

Other names

Depending on the field, recall goes by other names. In medical statistics it is called sensitivity, and its complement—the well-classified negatives—is specificity. In information retrieval one also speaks of recall: the proportion of relevant documents that a search manages to retrieve.