EvAnGCN: Evolving Graph Deep Neural Network Based Anomaly Detection in Blockchain

Detecting anomalous behaviors in the blockchain is important for maintaining its integrity. An imminent challenge is to capture the evolving model of transactions in the network. Representing the network with a dynamic graph helps model the system's time-evolving nature. However, as the graph evolves, real-world scenarios further stimulate the development of Graph Neural Networks (GNNs) to handle dynamic graph structures. In this paper, we propose a novel dynamic Graph Convolutional Network framework, namely EvAnGCN (Evolving Anomaly detection GCN), that helps detect anomalous behaviors in the blockchain. EvAnGCN exploits the time-based neighborhood feature aggregation of transactional features and the dynamic structure of the transaction network to detect anomalous nodes within the network. We conducted experiments on the Ethereum blockchain transaction dataset. Our experimental results demonstrate that EvAnGCH outperformed the baseline models.