Table 2.

Performance of LightGBM model for predicting mortality*

Target	AUROC	Precision	Recall	Accuracy	F1 score	Matthews correlation coefficient	Brier score
Training set
Top 5 features	0.81 (0.75–0.86)	0.59 (0.42–0.76)	0.38 (0.27–0.49)	0.74 (0.67–0.82)	0.44 (0.35–0.53)	0.45 (0.36–0.54)	0.06 (0.05–0.07)
Top 10 features	0.84 (0.78–0.90)	0.40 (0.29–0.52)	0.60 (0.49–0.71)	0.83 (0.77–0.89)	0.46 (0.38–0.54)	0.47 (0.41–0.53)	0.05 (0.04–0.06)
Top 15 features	0.87 (0.82–0.92)	0.49 (0.38–0.60)	0.56 (0.45–0.68)	0.85 (0.81–0.89)	0.49 (0.42–0.56)	0.49 (0.42–0.56)	0.04 (0.04–0.05)
Top 20 features	0.87 (0.82–0.92)	0.45 (0.38–0.53)	0.58 (0.41–0.75)	0.81 (0.75–0.87)	0.47 (0.41–0.53)	0.47 (0.41–0.53)	0.05 (0.04–0.05)
All features	0.84 (0.79–0.89)	0.47 (0.28–0.65)	0.53 (0.44–0.62)	0.82 (0.76–0.89)	0.44 (0.37–0.52)	0.45 (0.37–0.53)	0.05 (0.04–0.05)
Test set
Top 5 features	0.92	0.30	0.50	0.90	0.38	0.37	0.06
Top 10 features	0.93	0.40	0.67	0.89	0.50	0.50	0.03
Top 15 features	0.97	0.67	0.33	0.94	0.44	0.50	0.02
Top 20 features	0.96	0.36	0.67	0.92	0.47	0.48	0.03
All features	0.96	0.37	0.67	0.91	0.47	0.48	0.03

AUROC: Area under the receiver operating characteristic curve