Recall is a metric used in machine learning to evaluate the performance of a classification model. Especially in contexts where the detection of positive cases is crucial. It is particularly relevant in situations where missing a positive instance (e.g., failing to detect a true case of disease in medical diagnostics) is more critical than incorrectly labeling negative instances as positive.
Definition of Recall
Recall is defined as the ratio of the number of true positive predictions to the total number of actual positive instances in the dataset. It is calculated using the formula:
where:
- True Positives (TP) are the correctly identified positive cases.
- False Negatives (FN) are the positive cases that the model incorrectly identified as negative.
Understanding Recall
- A recall of 1 (or 100%) means that the model correctly identified all positive cases, but it does not mean that all predictions were correct. The model could have many false positives (negative instances incorrectly labeled as positive).
- A low recall indicates that the model missed a significant number of positive cases.
- Recall is particularly important in situations where the cost of missing a positive case is high (e.g., failing to diagnose a serious disease).
Recall in context
- Trade-off with Precision: Recall is often used in tandem with precision, another metric that measures the accuracy of the positive predictions made by the model. Optimizing for recall often decreases precision as the model tends to classify more instances as positive to ensure fewer positive cases are missed.
- Use in imbalanced datasets: In imbalanced datasets, where one class significantly outnumbers the other, recall becomes an important metric as it focuses only on the performance of the model in the minority class.
Example
In a medical test for a disease:
- If 100 patients actually have a disease (positives), and the model correctly identifies 90 of them as having the disease (TP), but misses 10 (FN), then the recall of the model is 90/100 = 0.9 or 90%.
Conclusion
Recall is a key metric in evaluating classification models, especially in scenarios where missing a positive instance carries significant consequences. It provides a clear measure of the model’s ability to detect all relevant instances, though it must be balanced with other metrics like precision to ensure a well-rounded evaluation of the model’s performance.
