Abstract
Accurate identification of cryptocurrency traffic is essential for regulatory and forensic applications such as anti-money laundering, illicit transaction tracing, and protocol-specific risk assessment. However, real-time classification remains challenging due to encryption heterogeneity across blockchain protocols and the interleaving of multiple currencies in peer-to-peer networks. To address this, we propose CryptoFormer, a Transformer-based framework for multi-currency traffic classification. CryptoFormer introduces a two-stage architecture that separately captures packet-level semantics and session-level behavioral patterns. An adaptive gated fusion module further integrates both content and behavioral features, enabling robust representation learning across plaintext and ciphertext protocols without decryption. We also construct a labeled real-world traffic dataset covering eighteen cryptocurrency assets and diverse background noise, capturing protocol diversity and deployment realism. Experimental results show that CryptoFormer achieves a macro F1-score of 95.02%, outperforming strong baselines including ET-BERT and TrafficFormer. Despite its shallow structure and smaller training corpus, CryptoFormer delivers comparable or better accuracy with significantly lower computational overhead, making it suitable for deployment in regulatory, compliance, and edge-monitoring systems.
| Original language | English |
|---|---|
| Journal | IEEE Transactions on Dependable and Secure Computing |
| DOIs | |
| Publication status | Accepted/In press - 2026 |
| Externally published | Yes |
Keywords
- Blockchain Governance
- Cryptocurrencies Identification
- Traffic Classification
- Transformer
Fingerprint
Dive into the research topics of 'CryptoFormer: Multi-Currency Traffic Classification Via Two-Stage Transformer Architecture'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver