Recall

TL;DR

Measures how many of the real positive cases the model correctly finds.
Computed from true positives and false negatives.
Especially important when missing positives (false negatives) is costly.

Definition

Recall is a metric used to evaluate the performance of a machine learning model in classification tasks. It is defined as the number of true positive predictions made by the model divided by the total number of positive instances in the test set.

\text{Recall} = \frac{\text{True positives}}{\text{True positives} + \text{False negatives}}

Explanation

Recall measures the proportion of actual positive examples that the model correctly identifies. True positives are instances correctly predicted as positive; false negatives are positive instances the model failed to identify. A higher recall indicates the model finds a larger share of the true positives. However, increasing recall can come at the cost of producing more false positives, so recall is often considered alongside other metrics such as precision and accuracy.

Examples

Example 1: Spam detection

Test set: 100 emails (20 spam, 80 not spam)

True positives: 15 (emails correctly identified as spam)
False negatives: 5 (emails missed as spam)
True negatives: 80 (emails correctly identified as not spam)
False positives: 0 (emails incorrectly identified as spam)

Calculation:

\text{Recall} = \frac{15}{15 + 5} = 75\%

This means the model correctly identified 75% of the spam emails in the test set.

Example 2: Cat detection in images

Test set: 1000 images (200 contain a cat, 800 do not)

True positives: 180 (images correctly identified as containing a cat)
False negatives: 20 (images missed as containing a cat)
True negatives: 800 (images correctly identified as not containing a cat)
False positives: 0 (images incorrectly identified as containing a cat)