Micro-averaged F1
Also known as: Micro F1, Frequency-weighted average F1
Micro-averaged F1 computes the F1-score by aggregating true positives, false positives, and false negatives across all classes, then calculating a single metric. It is equivalent to accuracy in multi-class classification and is useful when class distributions reflect their natural importance.
Key highlights
- Equivalent to overall accuracy, providing an intuitive interpretation in multi-class settings.
- Weights classes by their frequency, reflecting natural importance distribution.
- Computationally efficient to compute from aggregated confusion matrix.
- Useful for comparing overall performance without examining per-class metrics.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use micro-averaged F1 when class distributions are natural and reflect their importance, such as in standard balanced benchmarks. Use it when you want a single metric equivalent to overall accuracy that also balances precision and recall. Avoid micro F1 when classes are heavily imbalanced and you want equal importance for all classes; use macro F1 instead.
Strengths & limitations
- Equivalent to overall accuracy, providing an intuitive interpretation in multi-class settings.
- Weights classes by their frequency, reflecting natural importance distribution.
- Computationally efficient to compute from aggregated confusion matrix.
- Useful for comparing overall performance without examining per-class metrics.
- Biased toward majority classes in imbalanced datasets.
- Does not equally penalize poor performance on minority classes.
- Can mask poor performance on specific rare classes.
- Not appropriate for imbalanced datasets where all classes are important.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
Is micro-averaged F1 really the same as accuracy?
Yes, in multi-class classification, micro-averaged F1 is mathematically equivalent to overall accuracy. Both aggregate across all classes to compute a single metric. The difference is conceptual: accuracy is simpler, while micro F1 emphasizes the balance of precision and recall in this aggregate computation.
When should I use micro F1 instead of macro F1?
Use micro F1 when classes are balanced or their frequencies reflect their importance. Use macro F1 when all classes are equally important regardless of frequency, especially in imbalanced datasets. In imbalanced settings, macro F1 is usually preferable because it does not hide minority class problems.
Can macro and micro F1 differ significantly?
Yes. In imbalanced datasets, macro F1 can be much lower than micro F1 because macro F1 equally weights all classes while micro F1 emphasizes majority classes. This difference itself is informative: a large gap suggests class imbalance and uneven per-class performance.
Sources
- 1.Powers, D. M. (2011). Evaluation: From Precision, Recall and F-Measure to ROC, Informedness, Markedness and Correlation. Journal of Machine Learning Technologies, 2(1), 37-63.
- 2.Sokolova, M., Japkowicz, N., & Szpakowicz, S. (2006). Beyond Accuracy, F-Score and ROC: a Family of Discriminant Measures for Performance Evaluation. AI 2006, 4013, 1015-1021.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Micro-averaged F1. ScholarGate. https://scholargate.app/model-evaluation/micro-averaged-f1