Table 2.

Interlabel predictive performance of five ML models.

Models	label	Sensitivity	Specificity	PPV	NPV	Youden index	Accuracy
RF	0	0.333	0.980	0.750	0.891	0.313	0.831
	1	0.917	0.857	0.815	0.938	0.774	0.831
	2	0.962	0.909	0.893	0.968	0.871	0.932
GBDT	0	0.222	1.000	1.000	0.877	0.222	0.881
	1	0.875	0.829	0.778	0.906	0.704	0.847
	2	0.962	0.848	0.833	0.966	0.810	0.898
LR	0	0.222	0.900	0.286	0.865	0.122	0.797
	1	0.750	0.829	0.750	0.829	0.579	0.797
	2	0.961	0.909	0.893	0.968	0.871	0.932
AdaBoost	0	0.333	0.920	0.429	0.885	0.253	0.831
	1	0.792	0.857	0.792	0.857	0.649	0.831
	2	0.962	0.909	0.893	0.968	0.871	0.932
KNN	0	0.444	0.900	0.444	0.900	0.334	0.831
	1	0.750	0.886	0.818	0.838	0.636	0.831
	2	0.962	0.909	0.893	0.968	0.871	0.932

AUC, area under the curve; PPV, positive predictive value; NPV, negative predictive value; RF, random forest; GBDT, gradient boosting decision tree; LR, logistic regression; KNN, k-nearest neighbours.