Health insurance fraud detection by using an attributed heterogeneous information network with a hierarchical attention mechanism

Table 3 Anomaly detection effectiveness with balanced sample nodes

Train:Val:Test	1:1:3			2:1:2			3:1:1
Metrics	F1	Recall	Precision	F1	Recall	Precision	F1	Recall	Precision
Metapath2vec	0.795	0.7191	0.8888	0.8037	0.7166	0.9148	0.7384	0.6857	0.8
GCN	0.6067	0.6175	0.6296	0.5737	0.5908	0.5894	0.6036	0.6304	0.6121
HAN	0.7658	0.7762	0.8051	0.7543	0.7695	0.796	0.7896	0.8025	0.8138
MHAMFD_hierar	0.8014	0.8131	0.9022	0.8095	0.8658	0.9194	0.8662	0.8346	0.9143
MHAMFD_single	0.7611	0.7917	0.8908	0.7979	0.8391	0.8948	0.7942	0.8417	0.9021
MHAMFD_multi	0.7981	0.8014	0.8989	0.8048	0.8490	0.8748	0.7816	0.8392	0.8846
MHAMFD	0.8361	0.8764	0.9194	0.8679	0.8813	0.9386	0.8806	0.8692	0.9435

ISSN: 1472-6947