基于Zipf定律的随机森林财务预警模型  

Random Forest Financial Early Warning Model Based on Zipf's Law

作　　者：孟银凤[1] 王珮瑶 MENG Yinfeng;WANG Peiyao(School of Mathematical Sciences,Shanxi University,Taiyuan 030006,China)

机构地区：[1]山西大学数学科学学院,山西太原030006

出　　处：《山西大学学报（自然科学版）》2023年第4期821-829,共9页Journal of Shanxi University(Natural Science Edition)

基　　金：国家自然科学基金(61807022);山西省高等学校教学改革创新项目(J20220098)。

摘　　要：为了有效识别财务报表欺诈,解决现有欺诈检测模型应用局限性问题,并对上市公司财务危机做出准确预测,提出了一个基于Zipf定律的随机森林财务预警模型。首先,对财务数据集进行特征选择,运用随机森林算法计算特征变量重要性,筛选后得到用于分类检测的最优特征子集;其次,通过构造Zipf因子,生成新的特征向量补充到数据集中以挖掘更多有效信息,并将其与随机森林模型结合,识别具有财务欺诈风险的样例,得到分类预测结果;最后,基于1997—2017年中国A股上市公司的财务数据进行100次重复实验,以AUC值、准确率、召回率、特异度等指标对模型进行评价,并对比该模型与其他几种预警模型的预测性能,结果表明,基于Zipf定律的随机森林模型对上市公司财务风险的预测效果更优。In order to effectively identify fraud in financial statements,solve the application limitations of existing fraud detection models,and accurately predict the financial crisis of the listed companies,a random forest financial early warning model based on Zipf's law was proposed.Firstly,feature selection was carried out on the financial datasets,and the importance of features was calculated by using random forest algorithm to obtain the optimal feature subset for classification detection;secondly,Zipf factors were constructed to generate new features and added to the datasets to mine more effective information,which were combined with random forest model to identify examples with financial fraud risks and obtain classification prediction results;finally,based on the financial data from Chinese A-share listed companies from 1997 to 2017,100 repeated experiments were carried out to evaluate the model with AUC,accuracy,recall and specificity,and the prediction performance of the model was compared with that of other early warning models.The results show that the random forest model based on Zipf's law has better prediction effect on the listed company financial risks.

关 键 词：Zipf定律 财务预警 随机森林 财务欺诈 预测性能 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]