Table 4.

Comparison on predictive performance of three alternative machine learning classification algorithms using the same features under the full model

Algorithm	Full model* (based on 30 features)
	On Day 1 of admission
	Sensitivity		Specificity		Accuracy (%)
	Macro averaged (%)	Micro averaged (%)	Macro averaged (%)	Micro averaged (%)
Decision Tree #	76.1	90.4	78.3	90.4	90.4
Random forest #	77.8	90.9	69.6	90.9	90.9
As compared against the study’s chosen model
XGBoost	82.6	92.3	96.0	96.1	92.3

Algorithm	Full model* (based on 30 features)
	On Day 5 of admission
	Sensitivity		Specificity		Accuracy (%)
	Macro averaged (%)	Micro averaged (%)	Macro averaged (%)	Micro averaged (%)
Decision Tree #	91.3	97.1	98.0	97.1	97.1
Random forest #	92.3	97.6	95.3	97.6	97.6
As compared against the study’s chosen model
XGBoost	99.7	99.5	99.5	99.5	99.5

^*Model performance based on testing dataset (n = 208)

^#median imputation method was adopted to handle missing data values in study subjects