Table 3 Comparison of predictive performance of the five most commonly used machine learning models.

Dataset	Models	AUC(95%CI)	ACC	Sensitivity	Specificity	PPV	NPV
Training cohort	LR	0.901(0.845–0.956)	0.836	0.750	0.917	0.894	0.797
	Random Forest	0.942(0.903–0.981)	0.871	0.929	0.817	0.825	0.925
	XGBoost	0.985(0.969–0.999)	0.931	0.929	0.933	0.929	0.933
	SVM	0.901(0.845–0.957)	0.836	0.804	0.867	0.849	0.825
	MLP	0.845(0.777–0.913)	0.750	0.714	0.783	0.755	0.746
Internal validation cohort	LR	0.819(0.680–0.957)	0.830	0.750	0.871	0.750	0.871
	Random Forest	0.825(0.692–0.957)	0.809	0.687	0.871	0.733	0.844
	XGBoost	0.756(0.611–0.901)	0.745	0.750	0.742	0.6	0.852
	SVM	0.806(0.649–0.964)	0.83	0.812	0.613	0.520	0.864
	MLP	0.833(0.743–0.924)	0.738	0.710	0.929	0.985	0.325
External validation cohort	LR	0.817(0.695–0.940)	0.916	0.968	0.571	0.937	0.727
	Random Forest	0.666(0.514–0.818)	0.486	0.430	0.857	0.952	0.185
	XGBoost	0.638(0.474–0.801)	0.654	0.654	0.714	0.937	0.233
	SVM	0.815(0.696–0.934)	0.626	0.581	0.929	0.982	0.250
	MLP	0.778(0.652–0.903)	0.794	0.774	0.929	0.986	0.382

AUC area under the ROC curve, CI confidential interval, ACC accuracy, PPV positive predictive value, NPV negative predictive value, LR logistic regression, XGBoost extreme gradient boosting, SVM support vector machine, MLP multilayer perceptron.

Quick links

Search