Enhancing Anomaly Detection in Blockchain Transactions with Transformer-Based Models

Anomaly Detection (AD) in Bitcoin Transactions (BTXs) is critical for maintaining the integrity and security of blockchain systems. This study proposes a Transformer-based framework to address challenges such as class imbalance, complex feature interactions, and temporal dependencies inherent in Bitcoin Transaction (BTX) data. The model leverages self-attention mechanisms, feature embeddings, and positional encodings to capture intricate relationships among transaction attributes, enhancing its robustness and scalability. Comprehensive evaluations demonstrate the Transformer model’s superior performance, achieving 99.5% Accuracy, 94.0% Precision, 93.0% Recall, and 98.0% AUC-ROC, significantly outperforming traditional Machine Learning (ML) and Deep Learning (DL) approaches, including Random Forest (RF), Decision Tree (DT), Autoencoders, and Long Short-Term Memory (LSTM). Feature importance analysis identified key attributes such as total btc and mean in btc as primary contributors to AD, providing actionable insights for domain experts. The study highlights the Transformer model’s potential for real-time deployment in blockchain monitoring systems, paving the way for more secure and transparent decentralized networks. Future research could explore integrating additional contextual features and hybrid techniques to refine the framework’s applicability further.