Recall (Sensitivity)

Recall measures how well a model finds the positive cases that are actually present in the data. It is computed by taking the true positives, then comparing them to all real positives, which are the true positives plus the false negatives that the model missed. This works because every actual positive outcome must end up in one of those two groups: correctly detected or overlooked.

Use recall when missing a positive case is costly, such as screening for disease, fraud detection, or spam filtering where you want to catch as many real positives as possible. It is especially useful when comparing models that make different trade-offs between finding positives and raising false alarms.

• Using false positives in the denominator instead of false negatives
• Thinking recall measures overall accuracy rather than only coverage of actual positives
• Forgetting that recall depends only on actual positive cases, not on true negatives