An ML engineer is choosing a model for production where false negative predictions are far more costly than false positives. Which metric should the engineer prioritize most when selecting the model?

High recall is the most important metric when false negatives are costly. Recall measures the proportion of actual positive cases that are correctly identified by the model (True Positives / (True Positives + False Negatives)). Maximizing recall minimizes false negatives, which is crucial when missing a positive case has severe consequences. For example, in medical diagnosis, a false negative could mean a missed disease. Low precision would indicate many false positives, which are less costly in this scenario. High precision would mean fewer false positives, but doesn't directly address the cost of false negatives. Low recall would be detrimental, as it implies a high number of false negatives, directly contradicting the goal of minimizing them.