Medicare fraud detection using neural networks

Journal of Big Data

Table 14 Average ROS results (30 runs)

Method	\(n_{neg}{:}n_{pos}\)	Hidden layers	Decision threshold	ROC AUC	TPR	TNR	G-Mean
ROS-1-2	99:1	2	0.0110	0.8383	0.8572	0.6334	0.7338
ROS-1-4		4	0.0130	0.8325	0.8064	0.6857	0.7372
ROS-2-2	80:20	2	0.2410	0.8484	0.8282	0.6926	0.7549
ROS-2-4		4	0.3000	0.8440	0.8497	0.6165	0.7109
ROS-3-2	60:40	2	0.4080	0.8454	0.8056	0.7198	0.7582
ROS-3-4		4	0.4370	0.8438	0.8163	0.6820	0.7385
ROS-4-2	50:50	2	0.4530	0.8505	0.8084	0.7324	0.7692
ROS-4-4		4	0.4740	0.8389	0.8066	0.6861	0.7365
ROS-5-2	40:60	2	0.5630	0.8503	0.8163	0.7272	0.7701
ROS-5-4		4	0.5950	0.8423	0.8086	0.7023	0.7508