Precision recall choice
For a fraud detection model, would you prioritize precision or recall? Explain the trade-off.
Answers use simple, clear English.
Quick interview answer
Usually prioritize recall (catch more fraud) because false negatives are expensive, then tune threshold for acceptable precision/alert load. Show precision-recall curve and cost matrix. Accuracy is misleading under imbalance.
Detailed answer
Usually prioritize recall (catch more fraud) because false negatives are expensive, then tune threshold for acceptable precision/alert load. Show precision-recall curve and cost matrix. Accuracy is misleading under imbalance. In production, combine model score with rules and manual review queues.
Real example & use case
Threshold set for 95% recall; analysts review lower-precision band.
Pros & cons
Pros: aligns metric with business cost. Cons: high recall increases reviewer load.