Researchers Advocate MCC Over F1 And Accuracy
Chicco and Jurman publish a formal comment in PLOS Computational Biology on December 1, 2025, criticizing a tutorial by Faye Orcales et al. for recommending accuracy, F1 score, cross-validation, and reporting ROC AUC. They argue these metrics mislead on imbalanced genomics antibiotic-resistance datasets and recommend the Matthews correlation coefficient (MCC), reporting recall/precision/specificity/NPV, and repeated stratified 80/20 hold-out validation.
Key Points
- 1Advocate using the Matthews correlation coefficient (MCC) over F1 and accuracy for imbalanced binary classification.
- 2Show that F1 and accuracy give overoptimistic scores on skewed datasets, masking poor negative-class performance.
- 3Recommend reporting MCC plus recall, precision, specificity, and negative predictive value for reliable evaluation.
Scoring Rationale
Practical, high-credibility methodological guidance promotes MCC and repeated hold-out; limited novelty as these critiques are already established.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
