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

Table 3 Comparison between deep neural network, extremely gradient boosting and random forest based on various metrics in test dataset

	Accuracy	F1-measure	G-mean	MCC*	AUROC	AUPRC	Confusion matrix**
DNN	0.862	0.575	0.713	0.747	0.857	0.603	0.926	0.074
DNN	0.862	0.575	0.713	0.747	0.857	0.603	0.452	0.548
XGBoost	0.872	0.554	0.667	0.748	0.854	0.622	0.956	0.044
XGBoost	0.872	0.554	0.667	0.748	0.854	0.622	0.534	0.466
Random forest	0.869	0.543	0.659	0.741	0.840	0.578	0.955	0.045
Random forest	0.869	0.543	0.659	0.741	0.840	0.578	0.545	0.455

MCC Matthews Correlation Coefficient; AUROCReceiver Operating Characteristic Area Under Curve; AUPRC Precision-Recall Area Under Curve
^*MCC has been projected from [-1,1] to [0,1] by \(\frac{MCC+1}{2}\) formula
^**Predicted and actual, non-diabetic and diabetic percent are presented in confusion matrix

ISSN: 1472-6947