Skip to main content
Fig. 5 | BioData Mining

Fig. 5

From: The Matthews correlation coefficient (MCC) should replace the ROC AUC as the standard metric for assessing binary classification

Fig. 5

Heatmap of \(\bar{M}\) for the \(\text {TNR}+\text {TPR} > 1\) half of the ROC space. To get a global overview of the \(\bar{M}\) values as a function of \(\text {TNR}\) and \(\text {TPR}\) we display a heatmap representation using a yellow to red palette which highlights the non-linear behaviour of the mapping, as evidenced by the curved isolines. As a straightforward consideration, \(\bar{M}\) achieves high values only when both \(\text {TNR}\) and \(\text {TPR}\) are high: if one of the two rates is low, \(\bar{M}\) values are bounded into a medium range. As in the previous plot, due to the symmetry in the \(\text {MCC}_{\text {TNR},\text {TPR}}(p)\) equation, we can restrict the display to the case \(\text {TNR}+\text {TPR} > 1\)

Back to article page