Summary
- A new survey paper examines the fifty-year history, mathematical foundations, and practical implications of the Matthews Correlation Coefficient (MCC) in binary classification.
- The work clarifies long-standing misconceptions, formalizes MCC equivalences to Pearson's correlation and Chi2, and evaluates its behavior under class imbalance.
- The author introduces analytical derivatives and graphical contour plots that generalize to any confusion-matrix-based metric to reveal structural properties like bias and robustness.
Why it matters
Understanding the mathematical foundations and limitations of evaluation metrics like MCC helps machine learning practitioners select appropriate measures for robust and fair binary classification.