基于LightGBM的以太坊恶意账户检测方法  

Ethereum Malicious Account Detection Method Based on LightGBM

作　　者：边玲玉 张琳琳[1,2] 赵楷[1,2] 石飞[1,2] BIAN Lingyu;ZHANG Linlin;ZHAO Kai;SHI Fei(College of Information Science and Engineering,Xinjiang University,Urumqi 830046,China;College of Cyber Science and Engineering,Xinjiang University,Urumqi 830046,China)

机构地区：[1]新疆大学信息科学与工程学院,乌鲁木齐830046 [2]新疆大学网络空间安全学院,乌鲁木齐830046

出　　处：《信息网络安全》2020年第4期73-80,共8页Netinfo Security

基　　金：国家自然科学基金[61867006];新疆维吾尔自治区科技厅创新环境建设专项[PT1811];新疆维吾尔自治区高校科研计划[XJEDU2017T002,XJEDU2017M005];新疆维吾尔自治区创新环境(人才、基地)建设专项(自然科学基金)联合基金[2019D01C062,2019D01C041];国家级大学生创新计划[201910755030]。

摘　　要：由于区块链匿名性的特点,以太坊逐渐成为恶意账户利用漏洞攻击、网络钓鱼等手段实施欺诈的平台。针对上述问题,文章提出了一种基于Light GBM的以太坊恶意账户检测方法。首先通过收集并标注8028个以太坊账户,基于交易历史规律提取手工特征;然后使用自动特征构造工具featuretools提取统计特征;最后通过融合的两类特征训练Light GBM分类器完成以太坊恶意账户检测。实验结果表明,文章提出方法的F1值为94.9%,相较于SVM、KNN等方法更加高效准确,引入手工特征有效提升了恶意账户的检测性能。Due to the anonymity of the blockchain, Ethereum has gradually become a platform for malicious accounts to scam through vulnerabilities, phishing, and other methods. An Ethereum malicious account detection method based on LightGBM is proposed. By collecting and annotating 8028 Ethereum accounts, handcrafted features are extracted based on the history of transactions, and statistical features are extracted using featuretools. Finally, the LightGBM classifier is trained to detect malicious accounts in Ethereum through the fusion of two types of features. The experimental results show that the F1-Measure of the proposed method is 94.9%, which is more efficient and accurate than SVM, KNN and other methods. The introduction of handcrafted features can effectively improve the detection performance of malicious accounts.

关 键 词：区块链 恶意账户检测 以太坊 LightGBM 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]