【Author】
Patel, Vatsal; Pan, Lei; Rajasegarar, Sutharshan
【Source】NETWORK AND SYSTEM SECURITY, NSS 2020
【Abstract】Ethereum is one of the largest blockchain networks in the world. Its feature of smart contracts is unique among the other crypto-currencies and gained wider attention. However, smart contracts are vulnerable to attacks and financial fraud within the network. Identifying anomalies in this massive network is challenging because of anonymity. Using traditional machine learning-based techniques, such as One-Class Support Vector Machine and Isolation Forest are ineffective in Identifying anomalies in the Ethereum transactions because of its limitations in terms of capturing the internode or account relationship information in the transactions. Ethereum transactions can be effectively represented using an attributed graph with nodes and edges capturing the inter-dependencies. Hence, in this paper, we propose to use a One-Class Graph Neural Network-based anomaly detection framework for detecting anomalies in the Ethereum blockchain network. Empirical evaluation demonstrates that the proposed method is able to achieve higher anomaly detection accuracy than traditional non-graph based machine learning algorithms.
【Keywords】Ethereum blockchain; One-class methods; Graph neural networks
【摘要】以太坊是世界上最大的区块链网络之一。其智能合约的特点在其他加密货币中是独一无二的，并获得了广泛的关注。然而，智能合约很容易受到网络内的攻击和财务欺诈。由于匿名性，在这个庞大的网络中识别异常非常具有挑战性。使用传统的基于机器学习的技术，如单类支持向量机和隔离森林，在识别以太坊交易中的异常方面是无效的，因为其在捕获交易中的节点间或帐户关系信息方面存在局限性。以太坊交易可以使用带有节点和边缘的属性图有效地表示。因此，在本文中，我们提出使用一个基于单类图神经网络的异常检测框架来检测以太坊区块链网络中的异常。实验结果表明，该方法比传统的基于非图的机器学习算法具有更高的异常检测精度。
【关键词】以太坊区块链;看到下面成了一个方法;神经网络图
评论