Transformer-Based Threat Detection in Blockchain Healthcare Transactions

Blockchain-based healthcare systems offer enhanced security and data integrity, yet remain vulnerable to fraudulent transactions and unauthorized access. This study proposes a novel hybrid autoencoder-transformer model for anomaly detection in blockchain healthcare transactions, integrating graph convolutional networks (GCNs) for structured data representation, transformers for capturing long-range dependencies, and contrastive learning for improved fraud detection. The model was evaluated on a large-scale healthcare transaction dataset with over 1 million records. Experimental results demonstrate that the proposed approach achieves an accuracy of 94.7%, outperforming traditional machine learning models such as random forests (85.2%) and XGBoost (87.5%), as well as deep learning methods like LSTMs (89.1%) and CNNL-STMs (91.2%). Ablation studies highlight the importance of each component, with the removal of the GCN reducing accuracy by 2.9%. The model maintains an F1-score of 93.0% even under adversarial perturbations. Despite computational overhead, the proposed framework provides a robust and scalable solution for real-time threat detection in decentralized healthcare systems. Future work will explore self-supervised learning and adaptive transformer architectures to further improve model efficiency and generalizability.