Applied AI

Choose the errors that matter to the task

Editorial edition · 16 November 2026

The idea

Precision and recall describe different aspects of a selection. Precision asks how many selected items are relevant; recall asks how many relevant items were selected. Neither number explains the cost of a particular mistake without context.

Design choice

Choose a threshold with the downstream action in mind. Missing an optional recommendation differs from missing a safety-related review case. Report the class distribution and inspect actual false positives and false negatives alongside the summary score.

Worked example

Of ten selected records, eight are relevant: precision is 0.8. If sixteen relevant records exist in total, recall is 0.5. These answer different questions.

Before you ship

  • Define the positive class
  • Report both error types
  • Evaluate the chosen operating threshold

Examples are illustrative. Adapt the checks to your own environment.

Further reading

Model evaluation, scikit-learn

Search the library