Design And Application Of Financial Fraud Identification Model For Listed Companies Based On Data Mining Theory

In 2017 alone,many listed companies,such as Zhejiang Busen Apparel Co.,Ltd.and Anshan Heavy Mine Machinery Co.,Ltd.,were punished by the SFC for financial fraud.Financial fraud of listed companies has not only caused great losses to investors,but also damaged the function of resource allocation in the securities market.Therefore,it is particularly important to establish an effective financial fraud identification model to identify whether listed companies are financial fraud.In this paper,the A-share listed companies in manufacturing industry are selected as the research sample,and listed companies punished for financial statement fraud in 2009-2017 are selected as the fraud sample,and the listed companies which have not been punished by the regulatory authority are selected as the normal sample according to the 1:3 ratio.800 samples were used as training samples and 80 samples as test samples to test the accuracy of financial fraud identification model.According to the theory of fraud triangle,this paper classifies the fraud identification indicators and gets 54 fraud identification indicators.Considering that the relationship between the financial indicators of reaction pressure and the non-financial indicators of reaction opportunities and excuses may be linear or non-linear,this paper uses step-by-step method.Regression and principal component analysis were used to select indicators,and finally 14 indicators of fraud recognition were obtained.Financial fraud has the characteristics of small sample size compared with non-financial fraud companies,and financial fraud is often concerned by stakeholders.Aiming at this typical unbalanced data,this paper chooses BP neural network and Adaboost to build the model.In this paper,the accuracy of model identification is tested by testing samples,and it is concluded that the error rate of fraud identification model is 25% in the first category(misjudging financial crisis as financial normal),and 18.33% in the second category(misjudging financial normal as financial crisis).Finally,the audit model is applied to specific cases.It is concluded that the model has high popularization value.