数字货币交易所洗钱行为检测
Money Laundering Detection for Digital Currency Transactions
投稿时间:2021-06-23  修订日期:2021-09-08
DOI:
中文关键词:  数字货币交易  反洗钱  异常检测  特征建模  孤立点检测
英文关键词:digital currency transaction  anti-money laundering  anomaly detection  feature modeling  outlier detection
基金项目:国家自然科学基金资助项目(61872388, 62072470),国家重点研发计划(2018YFB1700403)
作者单位邮编
钟增胜 中南大学 计算机学院 410075
朱纯瑶 中南大学 计算机学院 
杨逸飞 中南大学 计算机学院 
廖忻橙 中南大学 计算机学院 
王任之 中南大学 计算机学院 
赵颖 中南大学 计算机学院 410075
周芳芳 中南大学 计算机学院 
施荣华 中南大学 计算机学院 
秦拯 湖南大学 信息科学与工程学院 
摘要点击次数: 24
全文下载次数: 0
中文摘要:
      数字货币交易中的洗钱行为区别于传统金融犯罪形态,具有强匿名,去中心化等特点,传统反洗钱技术手段难以直接适用,这给挖掘洗钱线索、鉴别洗钱行为和打击洗钱犯罪提出了新的挑战.针对数字货币交易所面对的洗钱行为检测需求和检测难点,本文构建了一个层次化加权的交易行为特征描述体系,提出了一个结合孤立点检测和小类簇检测的数字货币交易行为异常检测方法,实现了从交易行为到交易用户的洗钱可疑度量化度量.我们在真实数字货币交易所数据集上进行了评估实验,结果显示,异常交易行为和可疑洗钱用户的检测准确率分别为96.02%和95.05%,均优于基准算法.同时,本文算法的特征体系能对检测结果做出有效解释,帮助数字货币交易所安全员快速开展后续调查和取证工作.
英文摘要:
      Money laundering in digital currency exchanges is differentiated from traditional financial crimes due to its strong anonymity and decentralization. Existing anti-money laundering techniques cannot be directly applied to digital currency transactions. Considering the traceability, interpretability, and measurability of money laundering crime forensics, we design a four-stage money laundering detection approach: (1) Definition of transaction behavior. In this stage, we define a set of transactions of a user during a fixed period of time as a transaction behavior unit considering the combinability of money laundering. (2) Feature modeling of transaction behavior. In this stage, we construct a set of features to characterize transaction behaviors from the aspects of entry, exit, and currency-to-currency transactions. (3) Abnormal behavior detection. In this stage, we adopt outlier detection and small cluster detection to find out loud and subtle abnormal transactions. In regard to the measurability of money laundering crime forensics, we design a calculation method for abnormal degree values of transaction behaviors. (4) Suspicious user identification. In this stage, we analyze the suspicious score distributions of users and calculate a suspected-launderer score for each of them, based on which highly suspicious money launderers are provided to security officers along with their corresponding money laundering behaviors. A set of experiments are conducted on a real-world dataset. The results demonstrate that our approach can effectively detect highly-suspected transactions and traders, which can improve the efficiency of consequent digital investigations for anti-money laundering.
  查看/发表评论  下载PDF阅读器
关闭