Recall: true positives over all actual positives, TP / (TP + FN). It is the metric to optimise when a miss is the expensive error, as in fraud or disease screening. You can always buy recall by lowering the threshold, which is why it is quoted next to precision.
Card 1 of 259. Answer: Recall: true positives over all actual positives, TP / (TP + FN). It is the metric to optimise when a miss is the expensive error, as in fraud or disease screening. You can always buy recall by lowering the threshold, which is why it is quoted next to precision.
About · Quizzes · History of AI · To suggest new questions or report an error send me a dm.
Recall. It answers exactly that question: of all the examples that are actually positive, what fraction did the model label positive? Recall = TP / (TP + FN), where FN counts the positives the model missed. It is also called sensitivity, hit rate or the true positive rate.
Example: a test set holds 100 fraudulent transactions. The model flags 120 transactions, 80 of them real fraud. Recall is 80 / 100 = 0.80, because 20 frauds slipped through. Precision, the share of flagged transactions that are fraud, is 80 / 120 ≈ 0.67. The two metrics answer different questions from the same confusion matrix.
Recall matters most when a miss is costly: cancer screening, fraud, safety defects. It can always be pushed to 1.0 by predicting everything as positive, so it means little on its own. Interviewers expect you to pair it with precision, explain how lowering the threshold trades one for the other, and mention the F1 score or the precision-recall curve as ways to summarize both.
