Detection of Abnormal Bitcoin Transactions Based on Graph Convolutional Neural Networks

As cryptocurrencies like Bitcoin grow, detecting abnormal transactions is vital for security and preventing money laundering. Traditional methods fail to capture the complex relationships in Bitcoin's graph structure. This paper proposes a GCN-based approach, using a local hash attention mechanism to focus on key relationships and address data imbalance. With SMOTE for oversampling and Focal Loss for optimization, our method outperforms traditional techniques in precision, recall, and F1-score, offering a fresh approach to Bitcoin transaction security.