Detection of Fraudulent Transactions in the Cryptocurrency Market Using an Explainable Graph Isomorphism Network

Cryptocurrency markets are highly vulnerable to fraud due to the lack of comprehensive regulatory frameworks and the pseudonymous nature of transactions. This paper proposes a graph-based learning approach for detecting suspicious transactions in decentralized networks, with a focus on Bitcoin. We evaluate three graph neural network (GNN) models: Graph Convolutional Networks (GCN), Graph Attention Networks (GAT), and Graph Isomorphism Networks (GIN). Experiments on the Elliptic dataset, which contains highly imbalanced data between legitimate and illicit transactions, show that GIN outperforms other models in detecting fraudulent patterns. To enhance interpretability and provide insights into the decision-making process of the models, we integrate explainable AI techniques, specifically using GNNExplainer, which identifies the most influential nodes and features contributing to each fraud prediction. This approach not only enables analysts and regulators to understand why a transaction is classified as suspicious, improving trust and accountability in automated cryptocurrency fraud detection systems, but also achieves a high detection F۱-score of ۹۶.۲۲% for illicit transactions using GIN.

کلیدواژه ها:

Fraud Detection ، Graph Isomorphism Network ، Graph Neural Network ، Explainable AI ، Deep Learning ، Bitcoin