Recall, also called sensitivity or the true positive rate, measures how many actual positive cases a classification model correctly identifies. It focuses on finding positives and penalizes false negatives. section 14.6.2
Calculate recall as TP / (TP + FN), where TP is the number of correctly identified positives and FN is the number of missed positives. Compare it with specificity, which measures correctly identified negative cases. section 14.6.2
Predicting NBA playoff qualification
A logistic regression predicts whether NBA teams qualify for the postseason. The test results contain 40 true positives and 10 false negatives, so recall is 40 / (40 + 10) = 0.80. The model correctly identifies 80% of teams that actually qualify, while its specificity is 0.6333. section 14.6.2
Interpret recall alongside the consequences of classification errors:
- Prioritize recall when missing a positive case is especially costly, such as failing to detect an aggressive cancer. section 14.6.2
- Balance recall against specificity when false positives and false negatives matter similarly, as in NBA postseason classification. section 14.6.2
- Use a confusion matrix, accuracy, misclassification error, and ROC analysis to evaluate recall within a broader model assessment. section 14.6.2
