Machine learning-based mortality prediction models for smoker COVID-19 patients

Table 4 Performance results of top five “post-admission” models

Rank	Algorithm	Feature set	Parameters	Calibration	Accuracy	AUC	Precision	Recall	F1 Score	Log Loss	Brier Score
1	XGBoost	8	Colsample_bytree = 0.3 Learning _rate = 0.01 n_estimators = 300 max_depth = 15	Uncalibrated	0.909	0.952	0.921	0.894	0.904	0.323	0.090
1	XGBoost	8		Calibrated	0.905	0.951	0.921	0.885	0.899	0.246	0.072
2	XGBoost	5	Colsample_bytree = 0.3 Learning _rate = 0.1 n_estimators = 300 max_depth = 15	Uncalibrated	0.904	0.946	0.915	0.892	0.899	0.268	0.076
2	XGBoost	5		Calibrated	0.902	0.943	0.917	0.885	0.896	0.282	0.079
3	XGBoost	3	Colsample_bytree = 0.5 Learning _rate = 0.1 n_estimators = 300 max_depth = 8	Uncalibrated	0.901	0.945	0.904	0.897	0.897	0.282	0.080
3	XGBoost	3		Calibrated	0.897	0.944	0.904	0.886	0.892	0.288	0.081
4	XGBoost	7	Colsample_bytree = 0.5 Learning _rate = 0.01 n_estimators = 500 max_depth = 10	Uncalibrated	0.902	0.950	0.919	0.883	0.896	0.255	0.075
4	XGBoost	7		Calibrated	0.897	0.948	0.916	0.872	0.889	0.266	0.076
5	Ensemble	8	XGBoost, SVM, Random Forest, Decision Tree, KNN	Uncalibrated	0.891	0.936	0.885	0.899	0.890	0.326	0.094
5	Ensemble	8	XGBoost, SVM, Random Forest, Decision Tree, KNN	Calibrated	0.885	0.935	0.894	0.870	0.879	0.292	0.086

ISSN: 1472-6947