On the performance of Matthews correlation coefficient (MCC) for imbalanced dataset
Abstract
The Matthews Correlation Coefficient (MCC) is one of the popular measurements for classification accuracy. It has been generally regarded as a balanced measure which can be used even if the classes are of very different sizes. The study of this paper finds that this is not true. MCC deteriorates seriously when the dataset in classification are imbalanced. Experiment results and analysis show that MCC is not suitable for classification accuracy measurement on imbalanced datasets.
- Publication:
-
Pattern Recognition Letters
- Pub Date:
- August 2020
- DOI:
- 10.1016/j.patrec.2020.03.030
- Bibcode:
- 2020PaReL.136...71Z
- Keywords:
-
- Matthews correlation coefficient;
- Classification accuracy measurement;
- Performance evaluation;
- Imbalanced dataset