What Is Recall in Machine Learning?
Recall is a performance metric that measures how effectively an AI model identifies all the truly relevant instances in a dataset. In many business situations, the cost of missing something important is very high. For example, failing to detect a fraudulent transaction, a defective part on an assembly line, or a serious security threat can have severe consequences. The challenge is to quantify how good your system is at avoiding these costly misses.
How it helps#
Recall gives you a simple percentage score that shows how comprehensive your model's search is. A high recall score indicates that the model is successfully finding almost all of the items it's supposed to, minimizing the risk of overlooking critical events or information.
How it works#
Imagine you have 100 customer emails, and 10 of them are urgent support requests. You use an AI model to automatically flag these urgent emails, and it correctly identifies 8 of them, but it misses 2.
The recall is calculated by dividing the number of items correctly identified (8) by the total number of items that should have been identified (10). In this case, the recall is 8 / 10, or 80%. This tells you the model successfully "recalled" 80% of all the urgent emails.
How it is different#
Recall is focused on comprehensiveness—finding as many of the relevant items as possible. This is different from Precision, which measures the accuracy of the predictions that are made. Precision answers the question, "Of all the items we flagged as important, how many were actually important?" A high-recall system for medical screening might flag some healthy patients for a second look (lower precision) to ensure it doesn't miss any actual cases. In contrast, a high-precision system would only flag cases it is very sure about, but might miss some borderline ones.