<![CDATA[The rapid development of Decentralized Finance (DeFi) on the Ethereum blockchain has increased ERC20 token transaction volumes but also triggered a surge in anomalous activities such as Ice Phishing, NFT Scams, and Reward Scams. Conventional detection methods on tabular data often fail to capture complex relational patterns between accounts. This study proposes a Graph Neural Network (GNN) approach using the GraphSAGE architecture to detect anomalies in ERC20 transaction networks. Transaction datasets were acquired from Etherscan, processed into graphs with additional structural features (PageRank, degree), and imbalance was handled using class weighting. Based on testing using 33,848 transactions, the model yielded an average accuracy of 95.92% and a macro F1-score of 82.62%. These results prove that GraphSAGE is effective in learning feature representations and graph structures to detect various ERC20 fraud modes simultaneously.]]>
Copyrights © 2026
