GNN-based Detection of Malicious Transactions in Ethereum Networks: Leveraging Tornado Cash Transaction Patterns

GNN-based Detection of Malicious Transactions in Ethereum Networks: Leveraging Tornado Cash Transaction Patterns

연구 분야: Strategies

학회: BSCI '25: Proceedings of the 7th ACM International Symposium on Blockchain and Secure Critical Infrastructure

Detecting malicious transactions in blockchain networks such as Ethereum poses a significant challenge in security and anti-money laundering efforts. Tornado Cash, a decentralized mixer on Ethereum leveraging zero-knowledge proofs (zk-SNARKs), has been repeatedly abused to anonymize illicit funds and was officially sanctioned by the U.S. Office of Foreign Assets Control (OFAC) in 2022. In this study, we propose a Graph Neural Network (GNN)-based method to identify wallets involved in suspicious transactions by analyzing interaction patterns related to Tornado Cash. We define a new metric called Account Relevance to quantify transaction influence between wallets, and construct transaction graphs using a breadth first search(BFS) tracking algorithm. Leveraging a three layer Graph Convolutional Network (GCN), our model learns wallet representations and demonstrates high accuracy in detecting malicious actors. We evaluate our approach on real Ethereum transaction datasets encompassing Tornado Cash, decentralized finance (DeFi) platforms, centralized exchanges, and phishing-related accounts. Our model achieves up to 96.35% accuracy and 90.58% F1-score, demonstrating its effectiveness in detecting malicious wallets while maintaining robustness across diverse transaction types. This research contributes to advancing blockchain forensic techniques and supports regulatory and investigative efforts to combat financial crime in decentralized systems.