Skip to main content
. 2021 Feb 16;11:3932. doi: 10.1038/s41598-021-83141-z

Figure 3.

Figure 3

Model performance increases with lower predictive uncertainty. Shown are the area-under-the-curve (AUC) measures for receiver operator characteristics (ROC) analysis of sample predictions stratified by predictive uncertainty. The sample predictions with lowest uncertainty (p < 0.05) (blue curve, n = 72) achieved the highest performance (AUC = 0.86) compared to the entire grouped cohort irrespective of uncertainty (AUC = 0.83) (black curve, n = 95). Meanwhile, the sample predictions with greatest uncertainty (i.e., least certain predictions) showed the lowest classification accuracy (AUC = 0.5) (red curve, n = 23).