HeteroGCL: A Heterogeneous Graph Contrastive Learning Framework for Scalable and Sustainable Cryptocurrency AML

Anti-money laundering (AML) in cryptocurrency networks presents significant challenges due to complex transactional relationships, severe class imbalance, and limited labeled data, which severely constrain the scalability and label efficiency of existing AML systems. Traditional machine learning approaches treat transactions independently and fail to capture the intricate network structures inherent in money laundering schemes. To address these limitations, we propose HeteroGCL, a heterogeneous graph contrastive learning framework for scalable and sustainable cryptocurrency AML. Our approach models cryptocurrency transactions as a heterogeneous graph with multiple node and edge types and integrates a heterogeneous graph attention network with a graph contrastive learning module. By leveraging unlabeled data through topology-aware and attribute-aware graph augmentations, HeteroGCL mitigates label scarcity while enabling scalable and label-efficient AML model training while reducing reliance on costly manual annotation. Extensive experiments on the Elliptic dataset demonstrate that HeteroGCL achieves superior performance over state-of-the-art baselines, achieving an F1-score of 0.824 and an AUC of 0.912, with a 4.7% improvement in F1-score compared to the CARE-GNN baseline. The results indicate that the proposed framework effectively captures complex money laundering patterns while supporting scalable deployment of AML systems and improving the economic and operational sustainability of blockchain AML infrastructures.