Interpretable Machine Learning to Optimize Early In-Hospital Mortality Prediction for Elderly Patients with Sepsis: A Discovery Study

Table 3

Comparison of the performance of the six models.

Model	AUROC	Accuracy	Precision	score	Recall
XGBoost	0.871(95% CI: 0.854–0.888)	0.851 (95% CI: 0.842–0.863)	0.503 (95% CI: 0.461–0.611)	0.547 (95% CI: 0.539–0.611)	0.601 (95% CI: 0.585–0.701)
LGBM	0.870 (95% CI: 0.861–0.894)	0.876 (95% CI: 0.868–0.890)	0.654 (95% CI: 0.618–0.735)	0.467 (95% CI: 0.431–0.551)	0.363 (95% CI: 0.317–0.440)
LR	0.857 (95% CI: 0.845–0.881)	0.877 (95% CI: 0.867–0.884)	0.660 (95% CI: 0.627–0.703)	0.48 (95% CI: 0.416–0.525)	0.377 (95% CI: 0.309–0.419)
RF	0.807 (95% CI: 0.795–0.835)	0.866 (95% CI: 0.856–0.875)	0.632 (95% CI: 0.585–0.723)	0.362 (95% CI: 0.318–0.427)	0.253 (95% CI: 0.211–0.303)
DT	0.738 (95% CI: 0.706–0.791)	0.85 (95% CI: 0.836–0.857)	0.5 (95% CI: 0.455–0.555)	0.369 (95% CI: 0.363–0.433)	0.293 (95% CI: 0.272–0.354)
KNN	0.744 (95% CI: 0.682–0.806)	0.857 (95% CI: 0.840–0.875)	0.581 (95% CI: 0.426–0.717)	0.283 (95% CI: 0.254–0.312)	0.187 (95% CI: 0.107–0.267)