A Survey on Graph Neural Network Approaches for Fraud Detection in Blockchain Networks

Due to the fast growth of financial ecosystems supported by the blockchain technology, there is a corresponding increase in the amount of malicious activities like phishing attacks on the Ethereum platform or money laundering using Bitcoin. Standard approaches based on machine learning models and rules can no longer cope with such problems since their nature cannot be modeled using the relational and temporal graph structure that lies within blockchain transactions. Graph Neural Networks have proved themselves to be quite effective in this area by supporting aggregation processes on multi-hop neighborhoods as well as modeling both the attributes and structure of the graph. The aim of this paper is to present a survey on recent developments concerning GNNs for blockchain fraud detection, which have been published between 2022 and 2026. Moreover, several important issues like architecture types, namely homogeneous, temporal, heterogeneous, and hybrid ones, are discussed alongside with the class imbalance problem, for which focal loss and oversampling methods are considered. Finally gaps in research are highlighted, providing the directions for solving them.