Macro-averaged F1
Macro-averaged F1-Score · Also known as: Macro F1, Unweighted average F1
Macro-averaged F1 computes the F1-score independently for each class and then takes the unweighted arithmetic mean. It treats all classes equally, regardless of their frequency in the dataset, making it useful for imbalanced multi-class problems.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use macro-averaged F1 when all classes are important and you want equal weight regardless of frequency. It is essential for imbalanced multi-class problems where missing the minority class is critical. Use it instead of overall accuracy when class distributions are skewed and you want to penalize poor performance on rare classes equally with poor performance on common classes.
Strengths & limitations
- Weights all classes equally, preventing majority class dominance of the metric.
- Sensitive to poor performance on minority classes.
- Intuitive interpretation: represents average F1 across all classes.
- Suitable for imbalanced multi-class datasets.
- Rare classes with few samples may have highly variable F1-scores, making macro F1 unstable.
- Does not account for the actual number of samples in each class.
- Does not distinguish between different types of errors.
- Can give misleading results if some classes are much harder to classify than others.
Frequently asked
When should I use macro-averaged F1 instead of micro-averaged F1?
Use macro-averaged F1 when you want all classes to matter equally (appropriate for imbalanced datasets where minority classes are important). Use micro-averaged F1 when you want to weight by class frequency (appropriate when the natural distribution of classes reflects their importance). In imbalanced settings, macro F1 is usually preferable.
Why can macro-averaged F1 be unstable for rare classes?
Rare classes have few samples, so their F1-scores can fluctuate based on small changes in a few predictions. A single classification error on a rare class has a large effect on that class's F1-score and thus on the macro average. Stratified cross-validation helps stabilize estimates.
Can macro-averaged F1 be high while accuracy is low?
This can happen in severely imbalanced datasets. For example, if class A is 99% of the data and classes B and C are 0.5% each, a model could achieve high accuracy by predicting class A for everything, but macro F1 would be low because classes B and C have zero F1-score. This is actually desirable behavior when minority classes matter.
Sources
- 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. link ↗
- 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. DOI: 10.1007/11941439_114 ↗
How to cite this page
ScholarGate. (2026, June 3). Macro-averaged F1-Score. ScholarGate. https://scholargate.app/en/model-evaluation/macro-averaged-f1
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- F1-ScoreModel Evaluation↔ compare
- Micro-averaged F1Model Evaluation↔ compare
- Weighted F1Model Evaluation↔ compare