TABLE 4.

Model screening of prediction mortality COVID-19 patients shows a high AUC and specificity for most ML-based methods such as support vector machine, neural boosted and K Nearest Neighbors.

Method	N	Entropy R2	Misclassification rate	AUC	RASE	Generalized R2	Sensitivity	Specificity
Training set
Bootstrap forest	1224	0.419	0.125	0.927	0.290	0.532	63	97
Boosted tree	1224	0.414	0.109	0.917	0.287	0.527	73	97
Neural boosted	1224	0.312	0.143	0.874	0.316	0.416	61	95
Nominal logistic	1224	0.287	0.147	0.866	0.322	0.386	66	96
Generalized regression lasso	1224	0.269	0.152	0.863	0.325	0.365	74	97
Support vector machines	1224	0.248	0.147	0.885	0.322	0.34	85	99
Decision tree	1224	0.234	0.158	0.82	0.334	0.323	71	96
Fit stepwise	1224	0.22	0.152	0.825	0.335	0.308
K nearest neighbors	1224	0.152	0.168				80	98
Validation set
Neural boosted	519	0.291	0.144	0.857	0.317	0.392	57	94
Fit stepwise	519	0.282	0.146	0.846	0.319	0.382
Generalized regression lasso	519	0.258	0.152	0.844	0.325	0.353	72	97
Nominal logistic	519	0.240	0.152	0.840	0.328	0.331	61	94
Boosted tree	519	0.238	0.158	0.837	0.331	0.328	55	98
Decision tree	519	0.235	0.1541	0.8112	0.33513	0.3247	70	96
Support vector machines	519	0.2316	0.1387	0.8354	0.32522	0.3205	81	99
Bootstrap forest	519	0.1909	0.1638	0.8072	0.34281	0.2691	73	97
K nearest neighbors	519	0.0855	0.1734				83	96