F1 score is the harmonic mean of precision and recall for a classification model. Use it as one quantitative view of how effectively a classifier performs its job. section 7.3.1 section 8.6
Treat F1 score as a design and testing consideration rather than an isolated target. A strong numerical result may still be unsuitable when transparency, latency, cost, privacy, security, or production robustness matter. section 7.3.1
Use F1 score within a documented evaluation workflow:
- Compare F1 score with precision and recall instead of relying on accuracy alone. section 8.6
- Evaluate a range of performance measures to reflect what matters in the application. section 8.6
- Consider F1 score alongside nonfunctional tests and wider model-selection judgments. section 8.6
- Record the evidence and reasoning behind the final model choice, especially when metrics do not identify a clear winner. section 8.6
