Deep Learning
Recall is a performance metric that measures a classification model's ability to identify all relevant samples.
True
Proposed by @svpino
Card 1 of 259. Answer: True
True
Proposed by @svpino
Card 1 of 259. Answer: True
About · Quizzes · History of AI · To suggest new questions or report an error send me a dm.
True. Recall is the share of all relevant, or truly positive, samples that a model identifies: recall = TP / (TP + FN), where FN counts the relevant samples it missed. It is also called sensitivity or the true positive rate.
The word relevant comes from information retrieval, where precision and recall were first used to evaluate search systems. Say 40 documents in a collection are relevant to a query and the engine returns 25 results, 20 of them relevant. Recall is 20/40 = 0.5: half of what the user needed never appeared. Precision, the share of returned results that are relevant, is 20/25 = 0.8.
In classification, recall is computed per class. The negative class's recall is called specificity, and the mean of the per-class recalls is balanced accuracy. With 100 positives and 900 negatives, a model that catches 60 positives and clears 855 negatives has recall 0.6 and specificity 0.95: accuracy is a flattering 0.915, balanced accuracy only 0.775.