MCC Score – The only ML metric you need

The title might be a bit of a clickbait, but MCC (Matthews Correlation Coefficient) is a critical ML metric that every Data Scientist must know.

Metrics like the F1 score focus on only one class and its performance, but if you want a balanced model then you should be optimising your model on MCC score rather than on Accuracy or F1-score.

Let us take an example of –

y_true = [1,1,1,1,0,0,1, 0,1]
y_pred = [1,1,1,1,1,1,1, 0,0]

If we calculate the F1-score then it is ~0.77, but the MCC score is ~0.19, meaning that even though the model is very good at classifying the positive class, it is not very good at the negative class.

If we look at the formula for MCC –

\[MCC = \frac{(TPxTN)-(FPxFN)}{\sqrt{(TP+FP)(TP+FN)(TN+FP)(TN+FN)}}\] https://www.hostmath.com/Math/MathJax.js?config=OK

It should be clear that MCC gives equal focus on both TP and TN, since it is a correlation coefficient, its value ranges from -1 to 1.