A credit card fraud detection model must find as many fraudulent transactions (positives) as possible; only about 2% of transactions are fraud. Which two evaluation metrics should the data scientist optimize to prioritize capturing positives?

True positive rate (recall) measures the proportion of actual positive cases that were correctly identified. In fraud detection, this means identifying as many fraudulent transactions as possible, which is crucial when positives are rare. Area under the precision-recall curve (AUPRC) is particularly useful for imbalanced datasets, like fraud detection, because it focuses on the positive class. It evaluates the trade-off between precision (correctly identified positives among all predicted positives) and recall. Optimizing AUPRC ensures the model performs well in identifying positives while minimizing false positives. Specificity and false positive rate focus on the negative class (non-fraudulent transactions), which is not the primary concern here. Accuracy can be misleading in imbalanced datasets as a model that always predicts the majority class can still achieve high accuracy.