面向上市企业财务报表舞弊判断的机器学习算法研究  

Research on Machine Learning Algorithms for Fraud Judgment in Financial Statements of Listed Companies

作　　者：章银平 郭凤华[2] ZHANG Yin-ping;GUO Feng-hua(Anhui Finance and Trade Vocational College,Hefei 230601,Anhui,China;School of Accounting,Anhui University of Finance and Economics,Bengbu 233030,Anhui,China)

机构地区：[1]安徽财贸职业学院,安徽合肥230601 [2]安徽财经大学会计学院,安徽蚌埠233030

出　　处：《贵阳学院学报（自然科学版）》2022年第1期38-42,共5页Journal of Guiyang University：Natural Sciences

基　　金：2019年安徽省质量工程教师教学创新团队-会计专业创新教学团队(项目编号:2019cxtd090)。

摘　　要：随着我国经济的发展和社会的进步,一些上市公司为牟取私利而在财务报表上舞弊,严重扰乱了经济市场秩序,鉴于此,研究首先建立基于随机森林算法的财务舞弊判断模型,通过选择训练和测试样本,建立数学模型、设计舞弊评估方法,选择准确率、精确率、召回率和F1得分作为评价指标,对该模型判断财务舞弊的效果进行评估,并选择K邻近(KNN)、支持向量机(SVM)两种算法作为对照组。结果表明:随机森林模型准确率、精确率、召回率、F1得分和AUC值分别为0.681、0.031、0.712、0.048和0.732,均高于对照组两种算法,说明其对于舞弊样本和非舞弊样本识别能力都是最好的。With the economic development and social progress of China,some listed companies cheat on financial statements in order to seek private interests,which seriously disrupts the economic and market order.In view of this,this study proposes to apply machine learning algorithm to judge the fraud of financial statements of listed enterprises.Through the establishment of machine learning-based K-proximity algorithm model,support vector machine model and random forest model,and mathematical models to evaluate the effects of the three models in judging financial fraud,the accuracy,precision,recall and F1 scores are selected as evaluation index.The results show that the recall rate of the support vector machine model is the highest,reaching 71.2%;while the random forest model has advantages in other indicators.

关 键 词：机器学习 支持向量机 随机森林 财务舞弊 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]