非平衡数据集下基于XGBoost模型的财务舞弊识别研究  

Research on financial fraud identification based on XGBoost model in unbalanced datasets

作　　者：王琦 熊莎丽娜 詹柔 张露 杨鑫 张健 Wang Qi;Xiong Shaina;Zhan Rou;Zhang Lu;Yang Xin;Zhang Jian(School of Mathematics and Science,Southwest Forestry University,Kunming,Yunnan 650224,China)

机构地区：[1]西南林业大学数理学院,云南昆明650224

出　　处：《计算机时代》2023年第12期59-63,共5页Computer Era

基　　金：云南省教育厅科学研究基金项目“基于非线性逻辑回归的M-Score模型优化研究”(2022J0523);云南省高等学校大学生创新创业训练计划项目“基于数据挖掘的企业财务报表舞弊识别研究”。

摘　　要：针对现实中舞弊样本与非舞弊样本存在的数量不平衡情况,通过25个财务指标与2个非财务指标,运用过采样、欠采样技术及XGBoost模型进行财务报表舞弊识别研究。结果表明,SMOTE过采样方法与XGBoost模型的结合在非平衡数据集下具有较好的整体识别效果,对上市公司财务报表舞弊的智能识别有一定参考意义。In view of the unbalance in the number of fraud samples and non-fraud samples in reality,a study on financial statement fraud identification is conducted by applying over-sampling,under-sampling techniques and XGBoost model to 25 financial indicators and 2 non-financial indicators.The results show that the combination of SMOTE over-sampling method and XGBoost model has a good overall identification effect in the unbalanced dataset,which has certain reference significance for the intelligent identification of financial statement fraud of listed companies.

关 键 词：非平衡数据集 财务报表舞弊识别 SMOTE XGBoost 

分 类 号：TP311.1[自动化与计算机技术—计算机软件与理论] F275.5[自动化与计算机技术—计算机科学与技术]