Table 4.

Risk stratification based on the optimal cut-off value in the models.

Approaches	Patients (n = 4,006)	Probability		P-value
		Predicted	Actual
Logistic regression
Low risk (≤ 40.00%)	2,069	22.18%	23.88% (494/2,069)	<0.001
High risk (>40.00%)	1,937	77.19%	77.13% (1,494/1,937)
XGBoosting machine
Low risk (≤ 40.00%)	2,103	22.16%	23.92% (503/2,103)	<0.001
High risk (>40.00%)	1,903	78.06%	78.03 (1,485/1,903)
Random forest
Low risk (≤ 40.00%)	2,122	30.10%	24.18% (513/2,122)	<0.001
High risk (>40.00%)	1,884	69.43%	78.29% (1,475/1,884)
Gradient boosting machine
Low risk (≤ 40.00%)	2,110	22.37%	23.93% (505/2,110)	<0.001
High risk (>40.00%)	1,896	78.04%	78.22% (1,483/1,896)
Neural network
Low risk (≤ 40.00%)	2,144	19.01%	24.53% (526/2,144)	<0.001
High risk (>40.00%)	1,862	77.36%	78.52% (1,462/1,862)
Decision tree
Low risk (≤ 40.00%)	2,085	22.14%	24.36% (508/2,085)	<0.001
High risk (>40.00%)	1,921	77.62%	77.04% (1,480/1,921)

XGBooting, eXtreme Gradient Boosting.