macroaveraged

Macroaveraged refers to a method of computing evaluation metrics by first calculating them for each class separately and then averaging the results across all classes. In classification, macro averaging is often applied to precision, recall, and F1 score. The macroaveraged metric treats every class equally, regardless of how many instances belong to each class.

Calculation typically involves C classes. For each class c, compute the per-class metric p_c (precision), r_c (recall), or f1_c (F1 score) using the class's confusion matrix. The macroaveraged metric is then the arithmetic mean of these per-class values: macro P = (1/C) sum p_c, macro R = (1/C) sum r_c, macro F1 = (1/C) sum f1_c. For multi-label problems, the same principle applies by computing the metric for each label and averaging.

Macro averaging contrasts with micro averaging, which aggregates true positives, false positives, and false negatives across all classes before computing a single metric. Because macro averaging gives equal weight to all classes, it highlights performance on minority or rare classes and can reveal weaknesses that micro averaging might miss. However, it can be sensitive to classes with very few instances, leading to volatile averages when the class set is large or imbalanced.

Macroaveraged metrics are commonly used in multiclass and multi-label evaluation, especially when balanced performance across all classes is important. They are frequently reported alongside micro-averaged metrics to provide a fuller view of classifier behavior.