Development and validation of inpatient mortality prediction models for patients with hyperglycemic crisis using machine learning approaches

Table 3 Performance of different machine learning models for prediction of hyperglycemic crisis outcome

Criteria	LR	SVM	RF	RPART	XGBoost	MARS	NNET	AdaBoost
AUC	0.909	0.968	0.968	0.970	0.929	0.861	0.925	0.945
CI 95%(AUC)	0.832–0.972	0.945–0.986	0.947–0.987	0.935–0.987	0.879–0.969	0.755–0.965	0.877–0.960	0.904–0.973
ACC	0.958	0.970	0.970	0.952	0.970	0.964	0.952	0.976
CI 95%(ACC)	0.931–0.977	0.946–0.986	0.946–0.986	0.923–0.972	0.946–0.986	0.938–0.981	0.923–0.972	0.953–0.990
Kappa	0.610	0.746	0.746	0.626	0.767	0.681	0.611	0.797
Sensitivity	0.545	0.727	0.727	0.682	0.818	0.636	0.636	0.773
Specificity	0.987	0.987	0.987	0.971	0.981	0.987	0.974	0.990
PPV	0.750	0.800	0.800	0.625	0.750	0.778	0.636	0.850
NPV	0.968	0.981	0.981	0.977	0.987	0.975	0.974	0.984
F1	0.632	0.762	0.762	0.652	0.783	0.700	0.636	0.810

AUC: Area under curve; ACC: Accuracy; CI: Confidence interval; PPV: Positive predictive values; NPV: Negative predictive values

ISSN: 1472-6823