ANOM-DGCN: Detection of Anomalies in Dynamic Networks using Deviated Graph Convolution Network

In the digital era, the web and social networks have become an essential part of our society's daily lives. Generally, people worldwide use these networks to access or share information, but communication over these networks also has anomalous behavior. The anomalous behavior is the change in the network that is abnormal or rare occurrences that may relate to frauds, real-life events, shilling attacks, denial of service attacks, follower boosting etc. In this paper, we propose a novel method, i.e., ANOM-DGCN, which is modification of the graph convolution network. We use attributed graph that represents the network dynamics as attributes and communication as edges. We conduct experiments on publicly available datasets, such as Enron, DARPA, and TwitterSecurity, where our proposed method outperforms existing state-of-the-art models. ANOM-DGCN present results with AUC of 83% and provide spatial-temporal metadata for further analysis.