Table 7.

Comparisons between different models constructed using 24 h dataset in terms of their prediction accuracy.

Metric	Method	In-Hospital Mortality	Short-Term Mortality		Long-Term Mortality
			48 h	72 h	30 Days	1 Year
AUROC	CART	0.8049	0.8246	0.8064	0.8140	0.8511
	LR	0.8331	0.9014	0.8438	0.8434	0.8987
	MARS	0.8053	0.8843	0.8250	0.8102	0.8935
	RF	0.8623	0.9203	0.8705	0.8710	0.9096
	GB	0.8623	0.9249	0.8760	0.8736	0.9280
Specificity	CART	0.7639	0.7927	0.7537	0.7658	0.8197
	LR	0.7707	0.8313	0.7812	0.7684	0.8772
	MARS	0.6443	0.6151	0.6115	0.6054	0.6765
	RF	0.2520	0.3011	0.2056	0.2301	0.4212
	GB	0.8184	0.8507	0.8123	0.7882	0.9316
Sensitivity	CART	0.7402	0.8071	0.7578	0.7607	0.7983
	LR	0.7278	0.8090	0.7597	0.7683	0.7642
	MARS	0.8023	0.9137	0.8512	0.8409	0.8992
	RF	0.9899	0.9848	0.9932	0.9866	0.9580
	GB	0.7427	0.8325	0.7618	0.7821	0.7563
Precision	CART	0.2499	0.0905	0.2303	0.2425	0.0658
	LR	0.2658	0.1168	0.2673	0.2601	0.0955
	MARS	0.1933	0.0572	0.1756	0.1736	0.0423
	RF	0.1233	0.0348	0.1084	0.1122	0.0257
	GB	0.3030	0.1248	0.2831	0.2669	0.1495
F1-Statistic	CART	0.3736	0.1628	0.3532	0.3678	0.1216
	LR	0.3894	0.2041	0.3955	0.3887	0.1698
	MARS	0.3116	0.1077	0.2912	0.2878	0.0809
	RF	0.2193	0.0672	0.1955	0.2014	0.0500
	GB	0.4304	0.2171	0.4128	0.3980	0.2497

CART, classification and regression trees; LR, logistic regression; MARS, multivariate adaptive regression splines; RF, random forest; and GB, gradient boosting.