Classical ML, validation, metrics, interpretation, debugging · insight-1144draft — in review
Name the averaging mode of every multiclass F1 because micro-averaging equals accuracy
- Evidence type
- documented fact
- Difficulty
- easy
- Roles
- Data scientist, Data analyst
- Industries
- cross-industry
Summary: situation, action, reason, boundary
Situation
A multiclass report shows one F1 number for six ticket types, delay causes or defect classes.
Action
State the average mode, report per-class precision, recall and support, and choose macro when rare classes carry the cost, weighted to mirror volume.
Reason
Micro sums numerators and denominators over classes, which for single-label multiclass makes precision, recall and F1 all equal accuracy; macro and weighted change the weights.
Boundary
Multiclass and multilabel classification reports and model-comparison tables.
Symptom or risk
Micro-F1 is dominated by large classes and is identical to accuracy for single-label problems; macro-F1 treats a class of 20 rows like one of 20,000; readers compare unlike numbers.
Exceptions
In multilabel problems micro-averaging is not accuracy and is useful when a majority label should be downweighted.
Trade-offs
Macro averages are unstable when a class has a handful of rows; show support next to every class.
Pitfalls
Comparing a macro-F1 from one team with a weighted-F1 from another across model versions.
Lessons and concepts
Claims and their sources
In a multiclass setting with all labels included, micro-averaging produces precision, recall and F that are all identical to accuracy; macro gives equal weight to each class and weighted weights classes by their presence in the true data.