Classification
Balanced accuracy
Implemented
classification.balanced_accuracyDefinition
Mean of per-class recall. Reduces the influence of class imbalance compared with accuracy.
Formula
Balanced accuracy = (1/K) Σₖ TPₖ / (TPₖ + FNₖ)
Range: [0, 1]
Inputs and outputs
- y_true: array of class labels
- y_pred: array of predicted labels
Returns: float
Assumptions
No assumptions beyond valid, aligned inputs of the documented types.
Limitations
No metric-specific limitations are documented yet. Interpret the value alongside the task, data, and other metrics.
Python API
import evalsuite as es
es.balanced_accuracy(y_true, y_pred)References
- Brodersen, K. H., Ong, C. S., Stephan, K. E., & Buhmann, J. M. (2010). The balanced accuracy and its posterior distribution. Proceedings of the 20th International Conference on Pattern Recognition, 3121–3124.