Cross-chain Abnormal Transaction Detection via Graph-based Multi-model Fusion

Cross-chain technology accomplishes asset transfer and exchange between different blockchains through cross-chain transactions. The interoperability of cross-chain is along with anomalous activities. Due to the involvement of multiple parties and interactions in cross-chain transactions, they inherently possess the characteristics of a graph structure. Traditional methods for anomaly detection often overlook the interconnected nature of transactions and fail to extract efficient high-order features by graph structures. In this paper, we propose GMMCCT, a graph-based multi-model fusion approach for detecting abnormal cross-chain transactions. GMM-CCT integrates the LR-XGBoost-GCN-mixed model, and utilizes Node2vec to map nodes in the graph into a low-dimensional vector space. The extracted node features are used to achieve the classification of abnormal nodes in the graph. We consider five typical models to analyze the practicality of the proposed GMMCCT in the data testing. We implement a GMMCCT prototype system over a Multichain dataset, which contains 234,233 transactions. Experimental results demonstrate that GMMCCT achieves comparable performance with state-of-the-art single-chain schemes, with 82% precision and 89% recall for normal labels.