I am looking for a summary of machine learning questions and answers related to evaluation metrics. When should I prioritize precision over recall, or vice versa?
I am always afraid of mixing these up during an interview, especially when talking about imbalanced datasets. Does anyone have a good framework for choosing the right metric for a business-driven model?
Precision is prioritized when the cost of false positives is high, while recall is prioritized when the cost of false negatives—missing a critical event—is unacceptable.
1 answer
You should prioritize precision when the cost of a false positive is high, such as in spam detection where you do not want to flag legitimate emails, and prioritize recall when missing a positive case is dangerous, like in medical diagnostics where failing to catch a disease is catastrophic.
Think of it as a balance between safety and utility; the choice should always be driven by the specific business loss associated with the misclassification.
