Recall (Sensitivity): Measuring Positive Detection Rate
What You'll Learn
You'll learn how recall measures your model's ability to find all actual positive cases, a critical metric when missing positives is costly.
The Core Idea
Recall (also called sensitivity or true positive rate) answers the question: "Of all the actual positive cases, how many did my model successfully identify?"
While precision (from the previous lesson) focuses on the quality of your positive predictions, recall focuses on coverage — are you catching all the positives that exist?
The Formula
Recall = True Positives / (True Positives + False Negatives)
Or more simply:
Recall = True Positives / All Actual Positives
A Medical Analogy
Imagine a disease screening test. High recall means the test catches most people who actually have the disease — you're not missing many sick patients. A recall of 0.95 (95%) means you correctly identified 95 out of every 100 people who truly have the condition.
Low recall is dangerous when the cost of missing a positive is high. In cancer detection, missing a true cancer case (false negative) could be fatal, so you prioritize high recall even if it means some false alarms.
When Recall Matters Most
- Medical diagnosis: Missing a disease can be life-threatening
- Fraud detection: Failing to catch fraudulent transactions costs money
- Security screening: Missing a threat has severe consequences
Unlike accuracy, recall focuses specifically on the positive class and ignores how well you handle negatives. It's particularly valuable when you have imbalanced datasets where positives are rare but critical.
Key Takeaway: Recall measures what fraction of actual positives your model successfully detects — maximize it when missing positives is costly, even if it means more false alarms.