Diverse GNN Encoder-Decoder for Graph Anomaly Detection

Attributed graphs contain extra information on the vertices and edges. They can be used to model a plethora of real-world scenarios like social networks, bank transactions, and even academic citation data. Anomalies in such graphs can be irregularities or unusual patterns that are observed in the attributes or the structure of the graph. Anomaly detection in attributed networks is a crucial task, aiming to identify such anomalies. Existing methodologies use various deep learning techniques using graph neural networks, graph encoder-decoder architectures, and multi-layer perceptions. This study proposes a new approach to improve the existing methods using different types of neural networks. Additionally, it takes into account the different types of attributes (relations) on edges that are present in some datasets that contain real-world anomalies, like the DGraph-Fin. The experiments are performed on datasets with synthetic and real anomalies. We observed that the density of the graph affected the efficiency of the different types of Graph Neural Networks. We provide experimental results on four datasets and four different Graph Neural Network approaches and show that we have the same or improved precision and recall.