Recall
Recall, also known as sensitivity or the true positive rate, measures how well a classification model identifies actual positive cases. It is defined as the proportion of true positives out of all real positives. A high recall means the model misses very few positive instances.
Examples
- Medical testing: a highly sensitive COVID-19 test ensures nearly all infected patients are detected.
- Fraud detection: high recall guarantees most fraudulent transactions are flagged.
- Search engines: a search algorithm with high recall retrieves almost all relevant documents, though some irrelevant ones may appear.
Why it matters
- ✅ Key in high-risk domains where false negatives are unacceptable.
- ❌ High recall alone may lower precision, leading to many false positives.
- ⚖️ Best evaluated alongside precision, often combined in the F1-score.
Recall highlights the ability of a model to minimize false negatives, which is often more important than avoiding false positives in critical applications. For instance, in medical diagnostics, missing a disease case (false negative) could be life-threatening, whereas a false alarm may only require further testing.
The trade-off between recall and precision is central: boosting recall usually means relaxing decision thresholds, allowing more positive predictions, but at the cost of catching more irrelevant cases. Practitioners often tune this balance depending on domain requirements—for example, prioritizing recall in safety-critical tasks like cancer screening or disaster prediction, while prioritizing precision in legal document retrieval.
Technically, recall is derived directly from the confusion matrix (TP / (TP + FN)), and it is one of the key metrics visualized in ROC and Precision–Recall curves. Beyond simple measurement, it reflects an organization’s risk appetite—whether it is acceptable to miss positives or not.
📚 Further Reading
- Fawcett, T. (2006). An Introduction to ROC Analysis.
- Bishop, C. M. (2006). Pattern Recognition and Machine Learning.
