Medicare fraud detection using neural networks

Journal of Big Data

Table 15 Average ROS–RUS results (30 runs)

Method	Neg. Class Reduction (%)	\(n_{neg}{:}n_{pos}\)	Hidden layers	Decision threshold	ROC AUC	TPR	TNR	G-Mean
ROS–RUS-1-2	50	50:50	2	0.5090	0.8500	0.8029	0.7354	0.7665
ROS–RUS-1-4			4	0.4820	0.8454	0.8064	0.7189	0.7597
ROS–RUS-2-2	75	50:50	2	0.5218	0.8509	0.7876	0.7553	0.7710
ROS–RUS-2-4			4	0.5140	0.8443	0.7992	0.7175	0.7526
ROS–RUS-3-2	90	50:50	2	0.4850	0.8477	0.8104	0.7209	0.7625
ROS–RUS-3-4			4	0.5020	0.8425	0.8063	0.7161	0.7585