. 2021 Apr 28;23(4):e26075. doi: 10.2196/26075

Table 3.

Predictability of intensive care unit admissions per predictive model.

Model			F1 optimization, mean (SD)			F_β=2 (recall-oriented) optimization, mean (SD)
		F1-score		Recall	F1-score		Recall
At the time of hospitalization
	KNN^a	0.428 (0.039)		0.574 (0.156)	0.369 (0.037)		0.741 (0.088)
	DT^b	0.461 (0.016)		0.651 (0.056)	0.309 (0.083)		0.904 (0.033)^c
	RF^d	0.454 (0.027)		0.713 (0.088)^c	0.382 (0.103)		0.794 (0.128)
	XGB^e	0.505 (0.040)^c		0.541 (0.074)	0.431 (0.040)		0.766 (0.084)
	LR^f	0.250 (0.015)		0.622 (0.041)	0.248 (0.013)		0.651 (0.074)
	MLP^g	0.449 (0.060)		0.703 (0.145)	0.410 (0.039)		0.818 (0.050)
	LGBM^h	0.480 (0.023)		0.536 (0.048)	0.435 (0.025)^c		0.770 (0.051)
At the time of SARS-CoV-2 testing
	KNN	0.195 (0.016)		0.818 (0.090)	0.198 (0.007)		0.852 (0.039)
	DT	0.209 (0.012)		0.752 (0.135)	0.201 (0.007)		0.890 (0.036)
	RF	0.200 (0.008)		0.880 (0.038)	0.200 (0.008)		0.880 (0.044)
	XGB	0.205 (0.009)		0.857 (0.058)	0.203 (0.004)		0.914 (0.032)
	LR	0.200 (0.008)		0.847 (0.054)	0.201 (0.007)		0.880 (0.034)
	MLP	0.202 (0.006)		0.871 (0.049)	0.200 (0.008)		0.880 (0.037)
	LGBM	0.204 (0.012)		0.871 (0.074)	0.197 (0.012)		0.861 (0.066)

^aKNN: k-nearest neighbors.

^bDT: decision tree.

^cBest-performing models.

^dRF: random forest.

^eXGB: XGBoost.

^fLR: logistic regression.

^gMLP: multilayer perceptron.

^hLGBM: LightGBM.