Macro-Averaged Performance Measure

A Macro-Averaged Performance Measure is a multi-class classification performance measure that computes the unweighted arithmetic mean of per-class performance scores.

• AKA: Macro-Average Metric, Unweighted Mean Measure, Class-Averaged Measure.
• Context:
  • It can typically treat Macro-Averaged Class Contributions equally regardless of macro-averaged class frequency or macro-averaged class support.
  • It can typically compute Macro-Averaged Performance Scores by calculating individual macro-averaged per-class metrics and averaging without weighting.
  • It can typically emphasize Macro-Averaged Minority Class Performance as strongly as macro-averaged majority class performance.
  • It can typically reveal Macro-Averaged Model Weaknesses in handling macro-averaged rare classes that macro-averaged weighted metrics might obscure.
  • It can typically provide Macro-Averaged Fair Assessments when each macro-averaged class has equal importance in the macro-averaged application domain.
  • ...
  • It can often produce Macro-Averaged Lower Scores than micro-averaged performance measures in macro-averaged imbalanced datasets.
  • It can often be sensitive to Macro-Averaged Poor Performance in any single macro-averaged class.
  • It can often be combined with Macro-Averaged Standard Deviations to show macro-averaged performance variability across classes.
  • It can often be preferred in Macro-Averaged Research Settings where macro-averaged class balance is theoretically important.
  • ...
  • It can range from being a Simple Macro-Averaged Performance Measure to being a Complex Macro-Averaged Performance Measure, depending on its macro-averaged base metric complexity.
  • It can range from being a Stable Macro-Averaged Performance Measure to being a Volatile Macro-Averaged Performance Measure, depending on its macro-averaged class variance.
  • It can range from being a Binary-Derived Macro-Averaged Performance Measure to being a Native Multi-Class Macro-Averaged Performance Measure, depending on its macro-averaged computation approach.
  • It can range from being a Low Macro-Averaged Performance Measure to being a High Macro-Averaged Performance Measure, depending on its macro-averaged model quality.
  • ...
  • It can be calculated using Macro-Averaged Formulas: (1/n) × Σ(metric_i) for n classes.
  • It can be visualized through Macro-Averaged Performance Charts showing individual and averaged macro-averaged scores.
  • It can be compared with other Macro-Averaged Aggregation Methods to understand macro-averaged trade-offs.
  • ...
• Examples:
  • Macro-Averaged Precision-Recall Measures, such as:
    • Macro-Precision Measure for macro-averaged precision calculation.
    • Macro-Recall Measure for macro-averaged recall calculation.
    • Macro-F1 Measure for macro-averaged F1 score calculation.
  • Macro-Averaged Accuracy Measures, such as:
    • Macro-Accuracy Measure for macro-averaged per-class accuracy.
    • Macro-Balanced Accuracy Measure for macro-averaged sensitivity.
  • Macro-Averaged Probabilistic Measures, such as:
    • Macro-AUC Measure for macro-averaged ROC curve.
    • Macro-Log Loss Measure for macro-averaged probabilistic loss.
  • ...
• Counter-Example(s):
  • Micro-Averaged Performance Measure, which aggregates confusion matrix elements globally rather than averaging per-class scores.
  • Weighted-Averaged Performance Measure, which weights classes by sample support rather than treating equally.
  • Sample-Averaged Performance Measure, which averages over instances rather than over classes.
• See: Multi-Class Classification Performance Measure, Micro-Averaged Performance Measure, Weighted-Averaged Performance Measure, Macro-F1 Measure, Class Imbalance Problem, Arithmetic Mean, Per-Class Performance.

---