摘要
为了解决首次代币发行(ICO)欺诈检测研究中存在的特征建模单一、模型缺乏可解释性等问题,提出一种融合多源异构数据的ICO欺诈预测和可解释分析模型IICOFP。首先,融合ICO项目基本信息、评级分数、社交媒体等多源异构数据,通过Lasso特征选择和Tomek-Link欠采样更有效地实现对ICO的特征建模;其次,基于GBDT算法训练ICO欺诈预测模型,并引入SHAP框架从多个角度分析欺诈型ICO的影响因素,有力增强模型的可解释性。实验结果表明,该模型的准确率、精确率、召回率、F 1分数和AUC值分别达到87.76%、85.37%、90.52%、87.87%和87.82%,各项性能比已有的最佳模型提高了约2%~10%,验证了融合多源异构数据进行特征建模在ICO欺诈预测中的关键作用(实验数据及代码:https://github.com/Lujiarong1203/IICOFP)。
Aiming at the problems of single feature modelling and models lacking interpretability in the current ICO fraud detection research,this paper proposed an interpretable ICO fraud prediction model(IICOFP)by fusing multi-source heteroge-neous data.Firstly,it fused the ICO project basic information,rating scores,social media and other multi-source heterogeneous data,and effectively achieved feature modelling of ICOs by Lasso feature selection and Tomek-Link under-sampling.Secondly,it trained the ICO fraud prediction model based on the GBDT algorithm,and introduced the SHAP framework to ana-lyze the main influencing factors of fraudulent ICOs from multiple perspectives,which strongly enhanced the interpretability of the model.The experimental results show that the proposed IICOFP model achieves accuracy,precision,recall,F 1 score and AUC value of 87.76%,85.37%,90.52%,87.87%and 87.82%,respectively,and the performance of each of them is improved by about 2%~10%over the existing best model,which verifies the key role of fusing multi-source heterogeneous data for feature modelling in ICO fraud prediction(experimental data and code:https://github.com/Lujiarong1203/IICOFP).
作者
卢加荣
廖彬
刘怡
陈海龙
Lu Jiarong;Liao Bin;Liu Yi;Chen Hailong(College of Statistics&Data Science,Xinjiang University of Finance&Economics,rümqi 830012,China;College of Big Data Statistics,Guizhou University of Finance&Economics,Guiyang 550025,China;School of Public Health,Xinjiang Medical University,rümqi 830017,China)
出处
《计算机应用研究》
北大核心
2025年第2期357-364,共8页
Application Research of Computers
基金
新疆社会经济统计与大数据应用研究中心项目(XJEDU2024J100)。
作者简介
卢加荣(1992—),男,甘肃金昌人,讲师,博士研究生,主要研究方向为区块链、统计与机器学习;通信作者:廖彬(1986—),男,四川内江人,教授,博导,博士,主要研究方向为数据库技术、深度学习(liaobin665@163.com);刘怡(1993—),女,甘肃民勤人,硕士,主要研究方向为统计学、机器学习;陈海龙(1984—),男,甘肃陇西人,副教授,硕导,博士,主要研究方向为经济统计.
