ETX2Vec: Fraud detection algorithm for Ethereum using temporal biased random walk strategy
April 9, 2026
In the world of cryptocurrency, fraud detection is a critical component of maintaining the integrity of financial transactions. With the rise of Ethereum, a decentralized platform for creating smart contracts, the need for robust fraud detection algorithms has become increasingly apparent. In a recent study, researchers proposed a new algorithm called ETX2Vec (Ethereum Transactions to Vector) that aims to improve fraud detection within the Ethereum network by enhancing transaction subgraph construction and random walk strategies.
The complexity of the Ethereum transaction network poses unique challenges for fraud detection algorithms. Traditional graph embedding methods based on random walks often struggle to capture the dynamic nature of transactions and the flow of funds within the network. To address these shortcomings, the researchers behind ETX2Vec introduced several key innovations.
One of the primary enhancements introduced by ETX2Vec is a new approach to transaction subgraph construction. By extracting the first-order predecessor and successor neighboring nodes of a target node, the algorithm reconstructs transaction subgraphs that provide a comprehensive view of the flow of funds. This modification enables the random walk to better capture the intricacies of Ethereum transactions.
In addition to improving transaction subgraph construction, ETX2Vec also introduces novel random walk strategies. Two key advancements include selecting the next node based on the non-decreasing order of transaction timestamps, which captures the temporal dynamics of transactions, and designing a biased random walk strategy that takes into account both transaction timestamps and amounts. By introducing a parameter to control the weighting of these factors when calculating transition probabilities, ETX2Vec enhances the accuracy of fraud detection within the Ethereum network.
Experimental results have shown that ETX2Vec outperforms existing models in downstream node classification tasks, achieving an average performance of 96.04%. This performance improvement of 3.74% over the best model in similar studies highlights the effectiveness of ETX2Vec in understanding and processing Ethereum transactions. Notably, ETX2Vec even surpasses neural network models like GAT and GCN, showing its superiority in generating high-quality node embedding vectors.
In conclusion, the ETX2Vec algorithm represents a significant advancement in fraud detection within the Ethereum network. By leveraging innovative approaches to transaction subgraph construction and random walk strategies, ETX2Vec demonstrates superior performance in capturing the complex dynamics of Ethereum transactions. As the world of cryptocurrency continues to evolve, robust fraud detection algorithms like ETX2Vec will play a crucial role in ensuring the security and integrity of financial transactions within decentralized platforms like Ethereum.
