Table 4.

Diagnostic performance of different machine learning classifiers in internal validation cohort and external validation cohort

Model	AUC	Accuracy	Sensitivity	Specificity	PPV	NPV	F1 score
Internal validation cohort
XGBoost	0.89 (0.76–1.00)	0.86	0.83	0.97	0.94	0.72	0.87
LightGBM	0.87 (0.74–0.99)	0.84	0.83	0.91	0.94	0.69	0.88
Logistic	0.86 (0.69–1.00)	0.76	0.82	0.86	0.94	0.48	0.88
RandomForest	0.83 (0.66–0.98)	0.79	0.82	0.85	0.92	0.58	0.86
MLP	0.70 (0.50–0.90)	0.63	0.73	0.75	0.82	0.44	0.77
External validation cohort
XGBoost	0.84 (0.69–0.99)	0.79	1.00	0.50	0.64	1.00	0.87
LightGBM	0.77 (0.60–0.95)	0.61	0.61	0.60	0.73	0.46	0.67
Logistic	0.75 (0.56–0.94)	0.68	0.67	0.70	0.80	0.54	0.73
RandomForest	0.75 (0.56–0.93)	0.71	0.78	0.60	0.78	0.60	0.78
MLP	0.57 (0.33–0.79)	0.50	0.44	0.60	0.67	0.38	0.53

AUC, area under curve; LightGBM, light gradient boosting; MLP, multi-layer perceptron; NPV, negative predictive value; PPV, positive predictive value; XGBoost, eXtreme Gradient Boosting