Balanced Accuracy
Balanced Classification Accuracy · Also known as: Average Recall, Equal-weight Average Sensitivity
Balanced accuracy is the average of recall values computed for each class separately. It corrects for class imbalance by giving equal weight to the performance on each class, regardless of class frequency in the dataset.
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 balanced accuracy when dealing with imbalanced datasets where minority classes are important but underrepresented. It is essential in medical diagnosis, fraud detection, rare disease screening, and any domain where missing the minority class is costly. It provides a more honest assessment than overall accuracy when class distributions are skewed.
Strengths & limitations
- Automatically handles class imbalance by weighing each class equally, regardless of frequency.
- Prevents models from achieving high scores by simply predicting the majority class.
- Intuitive interpretation: represents average performance across all classes.
- Generalized easily to multi-class scenarios.
- Does not account for the cost of different types of errors (false positives vs. false negatives may have different consequences).
- Can mask performance on specific classes; a high balanced accuracy does not guarantee acceptable performance on all classes.
- May not be appropriate if classes have different importance weights in the application domain.
Frequently asked
How does balanced accuracy differ from accuracy in imbalanced datasets?
On imbalanced datasets, overall accuracy can be high while minority class performance is poor. Balanced accuracy weighs each class equally, so it reflects true performance across all classes. A model that achieves 90% accuracy but predicts 0% correctly on the minority class would have much lower balanced accuracy.
Is balanced accuracy the same as averaging precision and recall?
No. Balanced accuracy is the average of per-class recalls (one per class). Averaging precision and recall together would mix metrics with different definitions. Balanced accuracy focuses on recall because it measures how well each class is captured, independent of false positive rates.
When should I use balanced accuracy instead of F1-score or ROC-AUC?
Balanced accuracy is simpler and more interpretable than ROC-AUC, making it good for model selection. Use F1-score when you want to balance precision and recall. Use balanced accuracy when you want equal importance for all classes regardless of their size and the cost structure is symmetric.
Sources
- Brodersen, K. H., Ong, C. S., Stephan, K. E., & Buhmann, J. M. (2010). The balanced accuracy and its posterior distribution. 20th International Conference on Pattern Recognition (ICPR), 3121-3124. DOI: 10.1109/ICPR.2010.764 ↗
- 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 ↗
How to cite this page
ScholarGate. (2026, June 3). Balanced Classification Accuracy. ScholarGate. https://scholargate.app/en/model-evaluation/balanced-accuracy
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.
- AccuracyModel Evaluation↔ compare
- F1-ScoreModel Evaluation↔ compare
- Matthews Correlation CoefficientModel Evaluation↔ compare
- Recall (Sensitivity)Model Evaluation↔ compare
- SpecificityModel Evaluation↔ compare