Blockchain Payment Fraud Detection with a Hybrid CNN-GNN-LSTM Model

Blockchain payment fraud detection is hindered by extreme class imbalance, complex transaction relationships, and large-scale data. We propose a hybrid deep learning framework that integrates Convolutional Neural Networks (CNN), Graph Neural Networks (GNN), and Long Short-Term Memory (LSTM) networks to jointly model transaction features, graph topology, and temporal dynamics. The CNN extracts local feature interactions from transaction attributes, a Graph Attention Network (GAT) captures topological dependencies in the transaction graph, and a bidirectional LSTM models sequential behavior over time. A multi-head attention module adaptively fuses these complementary representations for robust classification. To support scalable training on large-scale graphs, we employ a Random Sample Partition (RSP) strategy. Experiments on the Elliptic Bitcoin dataset show strong performance, achieving ROC-AUC 0.975, PR-AUC 0.918, and F1-score 0.902, outperforming competitive graph- and sequence-based baselines. Ablation studies further confirm the contribution of each component. Overall, the proposed framework provides an accurate and scalable solution for detecting fraudulent activities in blockchain ecosystems.