Severe/critical COVID-19 early warning system based on machine learning algorithms using novel imaging scores

Table 3 Comparison prediction performances of different models

	Model	AUC (95%CI)	Accuracy	Sensitivity	Specificity	PPV	NPV	F1
Mean	XGBoost	0.841 (0.690-0.987)	0.745	0.768	0.88	0.92	0.572	0.823
SD	XGBoost	0.044 (0.066-0.018)	0.052	0.122	0.16	0.098	0.169	0.033
Mean	RandomForest	0.843 (0.709-0.974)	0.731	0.698	0.975	0.988	0.554	0.809
SD	RandomForest	0.081 (0.139-0.025)	0.105	0.145	0.05	0.025	0.147	0.097
Mean	logistic	0.892 (NaN-NaN)	0.828	0.88	0.848	0.932	0.666	0.896
SD	logistic	0.069 (NaN-NaN)	0.079	0.119	0.189	0.087	0.177	0.066
Mean	LightGBM	0.813 (0.690-0.937)	0.324	0.671	0.956	NaN	0.324	NaN
SD	LightGBM	0.067 (0.098-0.038)	0.071	0.092	0.058	NaN	0.071	NaN
Mean	MLP	0.617 (0.397-0.829)	0.655	0.634	0.764	NaN	0.503	NaN
SD	MLP	0.219 (0.228-0.212)	0.162	0.326	0.172	NaN	0.099	NaN

AUC: Area under the curve; PPV: Pulse pressure variation; NPV: Nucleopolyhedroviruses.