Diabetes mellitus risk prediction in the presence of class imbalance using flexible machine learning methods

Table 5 Comparison between various sampling methods on the performance of algorithms

	Accuracy	F1-measure	G-mean	MCC	AUROC	AUPRC
DNN
RENN	0.830	0.583	0.773	0.745	0.862	0.608
OSS	0.856	0.594	0.747	0.753	0.855	0.599
SMOTE	0.805	0.556	0.768	0.729	0.855	0.594
SVM-SMOTE	0.827	0.580	0.773	0.743	0.856	0.602
ENN-SMOTE	0.818	0.563	0.763	0.733	0.850	0.599
XGBoost
RENN	0.814	0.572	0.779	0.740	0.856	0.588
OSS	0.831	0.554	0.733	0.727	0.842	0.591
SMOTE	0.859	0.568	0.708	0.742	0.848	0.592
SVM-SMOTE	0.844	0.555	0.718	0.730	0.858	0.605
ENN-SMOTE	0.857	0.548	0.688	0.733	0.845	0.594
Random forest
RENN	0.808	0.556	0.764	0.728	0.844	0.553
OSS	0.832	0.561	0.741	0.731	0.837	0.569
SMOTE	0.842	0.543	0.704	0.724	0.840	0.550
SVM-SMOTE	0.838	0.548	0.717	0.726	0.842	0.552
ENN-SMOTE	0.844	0.531	0.687	0.719	0.843	0.541

ISSN: 1472-6947