融合财务、管理和文本指标的财务欺诈识别研究  

Research on Financial Fraud Identification of Integrating Financial,Management and Text Indicators

作　　者：赵盛喆 李诗轩 姜慜喆 康金鑫 ZHAO Shengzhe;LI Shixuan;JIANG Minzhe;KANG Jinxin(School of Safety Science and Emergency Management,Wuhan University of Technology,Wuhan 430070,China;不详)

机构地区：[1]武汉理工大学安全科学与应急管理学院,湖北武汉430070 [2]东南大学经济管理学院,江苏南京210096

出　　处：《武汉理工大学学报（信息与管理工程版）》2024年第5期766-772,共7页Journal of Wuhan University of Technology：Information & Management Engineering

基　　金：国家自然科学基金青年项目(72204194)。

摘　　要：财务欺诈行为不仅会损害上市公司股东的经济利益,还会扰乱资本市场的正常运行,因此构建准确高效的财务欺诈识别模型尤为重要。选取2015—2021年间存在财务欺诈行为的上市公司,根据融合财务、管理和文本指标,构建财务欺诈识别指标体系,运用SMOTE方法处理非平衡数据集,XGBoost算法筛选关键指标,并分析了单分类器、集成学习和深度学习模型在财务欺诈方面的识别特性。研究发现:不仅是财务指标,管理指标和文本指标也对识别财务欺诈产生影响,且文本指标贡献大于管理指标;RNN是表现最好的模型,在所有特征融合后,准确率、AUC和F值分别达到了90.58%、93.34%和89.76%。Financial fraud not only harms the economic interests of shareholders of listed companies but also disrupts the normal operation of the capital market.Therefore,it is particularly important to build an accurate and efficient financial fraud identification model.This study selected listed companies that exhibit financial fraud between 2015 and 2021,integrated financial indicators,management indicators,and text indicators,and constructed a financial fraud identification indicator system.The SMOTE method was used to process imbalanced datasets,the XGBoost algorithm was utilized to screen key indicators,and the identification characteristics of single classifiers,ensemble learning,and deep learning models in financial fraud were analyzed.The research has found that not only financial indicators but also management indicators and text indicators are impactful in identifying financial fraud,and text indicators contribute more to identifying financial fraud than management indicators do.RNN is the best-performing model,with accuracy,AUC and F values reaching 90.58%,93.34%,and 89.76%,respectively,after fusing all features.

关 键 词：财务欺诈识别 文本指标 XGBoost 上市公司 深度学习 

分 类 号：G356[文化科学—情报学]