Table 4.

Best test prediction results of stacked machine-learning algorithms for one-versus-rest classification.

	Mild	Moderate	Severe
ROC	0.86	0.81	0.95
Accuracy, % (95% CI)	83.6 (78.7–87.8)	78.8 (73.5–83.5)	90.9 (86.8–94.0)
Sensitivity, %	86.1	60.0	73.6
Specificity, %	81.0	84.7	97
Positive predictive value, %	81.9	54.9	89.8
Negative predictive value, %	85.4	87.2	91.2

ROC, receiver operating characteristics; CI, confidence interval.