In multilabel classification, why can accuracyscore be low even if most individual labels are correct?
Classic MLLevel: IntermediateExact multilabel accuracy
In multilabel classification, why can accuracy_score be low even if most individual labels are correct?
Hints
- Is the unit a label or a complete label vector?
- Exact matching is stricter than per-label matching.
Answer
It uses subset accuracy: a sample counts as correct only when the entire predicted label set matches. One incorrect label makes that sample incorrect.
Source and licence: scikit-learn 1.7.2 — model evaluation · BSD-3-Clause · answer: Quantoros editorial