Precision and Recall

Category: science

Two metrics used to assess the accuracy of a classification model.

Precision is the "quality" (when the AI says it’s fraud, how often is it right?). Recall is the "coverage" (how much of the total fraud did the AI find?). You have to pick your poison: a high-precision model might miss some fraud, but it never triggers unnecessary investigations.

Common Examples

• We optimized for recall to ensure that we capture 100% of the severe liability risks, even if it slightly increases our false-positive review load.
• There is always a trade-off between precision and recall; our agency’s goal is to maximize the balance for our specific business needs.