Micro-Averaged Performance Measure

A Micro-Averaged Performance Measure is a multi-class classification performance measure that aggregates confusion matrix elements globally before computing performance metrics.

• AKA: Micro-Average Metric, Global Aggregation Measure, Pooled Performance Measure.
• Context:
  • It can typically aggregate Micro-Averaged Confusion Matrix Elements including micro-averaged true positives, micro-averaged false positives, and micro-averaged false negatives across all classes.
  • It can typically weight Micro-Averaged Class Contributions implicitly by their micro-averaged sample count, giving more influence to micro-averaged frequent classes.
  • It can typically provide Micro-Averaged Global Performance that reflects overall micro-averaged system accuracy across the entire micro-averaged dataset.
  • It can typically favor Micro-Averaged Majority Class Performance in micro-averaged imbalanced datasets due to micro-averaged sample-based weighting.
  • It can typically equal Micro-Averaged Overall Accuracy in certain micro-averaged single-label settings when computed appropriately.
  • ...
  • It can often produce Micro-Averaged Higher Scores than macro-averaged performance measures when micro-averaged majority classes perform well.
  • It can often be preferred when Micro-Averaged Overall System Performance matters more than micro-averaged per-class fairness.
  • It can often be interpreted as Micro-Averaged Instance-Level Performance rather than micro-averaged class-level performance.
  • It can often be computed efficiently from Micro-Averaged Global Counters without per-class calculation.
  • ...
  • It can range from being a Simple Micro-Averaged Performance Measure to being a Complex Micro-Averaged Performance Measure, depending on its micro-averaged metric complexity.
  • It can range from being a Sparse Micro-Averaged Performance Measure to being a Dense Micro-Averaged Performance Measure, depending on its micro-averaged prediction distribution.
  • It can range from being a Consistent Micro-Averaged Performance Measure to being a Variable Micro-Averaged Performance Measure, depending on its micro-averaged temporal stability.
  • It can range from being a Low Micro-Averaged Performance Measure to being a High Micro-Averaged Performance Measure, depending on its micro-averaged classification quality.
  • ...
  • It can be calculated using Micro-Averaged Formulas that pool micro-averaged counts before metric computation.
  • It can be implemented in Micro-Averaged Evaluation Frameworks for micro-averaged model comparison.
  • It can be tracked through Micro-Averaged Learning Curves during micro-averaged training processes.
  • ...
• Examples:
  • Micro-Averaged Precision-Recall Measures, such as:
    • Micro-Precision Measure for micro-averaged global precision.
    • Micro-Recall Measure for micro-averaged global recall.
    • Micro-F1 Measure for micro-averaged global F1 score.
  • Micro-Averaged Classification Measures, such as:
    • Micro-Accuracy Measure for micro-averaged overall correctness.
    • Micro-Specificity Measure for micro-averaged true negative rate.
  • Micro-Averaged Probabilistic Measures, such as:
    • Micro-AUC Measure for micro-averaged ROC analysis.
    • Micro-Brier Score for micro-averaged calibration.
  • ...
• Counter-Example(s):
  • Macro-Averaged Performance Measure, which computes unweighted average of per-class metrics rather than global aggregation.
  • Weighted-Averaged Performance Measure, which applies explicit class weights rather than implicit sample-based weighting.
  • Sample-Wise Performance Measure, which evaluates each instance individually rather than aggregating globally.
• See: Multi-Class Classification Performance Measure, Macro-Averaged Performance Measure, Weighted-Averaged Performance Measure, Micro-F1 Measure, Confusion Matrix, Global Aggregation, Instance-Level Evaluation.

---