Joint modeling strategy for using electronic medical records data to build machine learning models: an example of intracerebral hemorrhage

Tang, Jianxiang; Wang, Xiaoyu; Wan, Hongli; Lin, Chunying; Shao, Zilun; Chang, Yang; Wang, Hexuan; Wu, Yi; Zhang, Tao; Du, Yu

doi:10.1186/s12911-022-02018-x

BMC Medical Informatics and Decision Making

Table 4 The average performance of 7 models under each resampling method

From: Joint modeling strategy for using electronic medical records data to build machine learning models: an example of intracerebral hemorrhage

	Resampling	Models							Average
	Resampling	LR	RF	ANN	SVM	KNN	Stacking	AdaBoost	Average
Acc*	Original	0.792	0.812	0.812	0.811	0.797	0.815	0.815	0.808
	RUS	0.786	0.801	0.799	0.790	0.795	0.794	0.800	0.795
	ROS	0.784	0.796	0.796	0.781	0.780	0.790	0.794	0.789
	ADASYN	0.768	0.785	0.777	0.769	0.777	0.781	0.782	0.777
	BSMOTE*	0.751	0.766	0.767	0.758	0.756	0.762	0.763	0.760
	SMOTEENN	0.738	0.732	0.733	0.738	0.744	0.727	0.730	0.735
	Average	0.770	0.782	0.781	0.774	0.775	0.778	0.781	0.777
Sen*	Original	0.408	0.425	0.426	0.428	0.382	0.436	0.443	0.421
	RUS	0.493	0.466	0.471	0.492	0.469	0.484	0.464	0.477
	ROS	0.499	0.477	0.489	0.511	0.452	0.499	0.479	0.487
	ADASYN	0.470	0.480	0.484	0.500	0.480	0.492	0.480	0.484
	BSMOTE*	0.514	0.501	0.502	0.511	0.496	0.520	0.504	0.507
	SMOTEENN	0.534	0.540	0.541	0.545	0.504	0.553	0.545	0.537
	Average	0.486	0.482	0.486	0.498	0.464	0.497	0.486	0.486
Spe*	Original	0.962	0.984	0.985	0.982	0.982	0.984	0.981	0.980
	RUS	0.916	0.952	0.947	0.924	0.942	0.933	0.950	0.938
	ROS	0.911	0.940	0.933	0.901	0.927	0.920	0.935	0.924
	ADASYN	0.902	0.921	0.908	0.889	0.909	0.911	0.917	0.908
	BSMOTE*	0.857	0.885	0.887	0.869	0.871	0.871	0.879	0.874
	SMOTEENN	0.829	0.819	0.819	0.826	0.851	0.805	0.812	0.823
	Average	0.896	0.917	0.913	0.898	0.914	0.904	0.912	0.908
AUC	Original	0.728	0.740	0.748	0.736	0.730	0.740	0.745	0.738
	RUS	0.738	0.749	0.740	0.737	0.734	0.734	0.736	0.738
	ROS	0.736	0.751	0.750	0.744	0.734	0.747	0.743	0.744
	ADASYN	0.729	0.738	0.732	0.732	0.715	0.716	0.719	0.726
	BSMOTE*	0.734	0.740	0.733	0.734	0.720	0.730	0.730	0.732
	SMOTEENN	0.726	0.721	0.724	0.727	0.705	0.692	0.719	0.716
	Average	0.732	0.740	0.738	0.735	0.723	0.726	0.732	0.732
Pre*	Original	0.835	0.927	0.928	0.920	0.905	0.925	0.912	0.907
	RUS	0.730	0.831	0.821	0.786	0.787	0.801	0.825	0.797
	ROS	0.724	0.798	0.800	0.743	0.738	0.776	0.789	0.767
	ADASYN	0.691	0.773	0.755	0.737	0.733	0.764	0.765	0.745
	BSMOTE*	0.638	0.736	0.733	0.724	0.701	0.728	0.729	0.713
	SMOTEENN	0.640	0.691	0.686	0.698	0.681	0.684	0.692	0.682
	Average	0.710	0.793	0.787	0.768	0.758	0.780	0.785	0.769
F1*	Original	0.546	0.581	0.582	0.582	0.535	0.591	0.595	0.573
	RUS	0.585	0.590	0.591	0.591	0.585	0.591	0.587	0.589
	ROS	0.587	0.591	0.596	0.590	0.558	0.593	0.589	0.586
	ADASYN	0.554	0.579	0.574	0.574	0.571	0.582	0.577	0.573
	BSMOTE*	0.558	0.573	0.575	0.571	0.558	0.577	0.571	0.569
	SMOTEENN	0.560	0.561	0.563	0.570	0.553	0.564	0.563	0.562
	Average	0.565	0.579	0.580	0.580	0.560	0.583	0.580	0.575

*BSMOTE: Borderline SMOTE; Acc: Accuracy; Sen: Sensitivity; Spe: Specificity; Pre: Precision; F1: F1 score; Bold indicates the best value

Back to article page

ISSN: 1472-6947

Contact us

General enquiries: journalsubmissions@springernature.com