Principle:Scikit learn contrib Imbalanced learn Imbalanced Classification Reporting
| Knowledge Sources | |
|---|---|
| Domains | Machine_Learning, Model_Evaluation, Imbalanced_Learning |
| Last Updated | 2026-02-09 03:00 GMT |
Overview
A comprehensive evaluation report that compiles imbalance-specific metrics (precision, recall, specificity, f1, geometric mean, and IBA) per class and overall for classifiers on imbalanced datasets.
Description
Imbalanced classification reporting extends sklearn's classification_report with metrics specifically designed for imbalanced evaluation. The report includes per-class precision, recall (sensitivity), specificity, f1-score, geometric mean, index balanced accuracy (IBA), and support. This provides a complete picture of classifier behavior across all classes, highlighting both majority and minority class performance.
Usage
Use this principle as the final evaluation step after training a classifier on imbalanced data. The report replaces sklearn's standard classification_report with imbalance-aware metrics.
Theoretical Basis
For each class i:
- Precision: TP_i / (TP_i + FP_i)
- Recall (Sensitivity): TP_i / (TP_i + FN_i)
- Specificity: TN_i / (TN_i + FP_i)
- F1: 2 * precision * recall / (precision + recall)
- Geometric Mean: sqrt(sensitivity * specificity)
- IBA: (1 + alpha * dominance) * gmean^2