Research On Financial Fraud Identification Of Chinese Listed Companies Based On Data Mining

At present,China market economy has a good prospect,the development momentum is swift and violent.The number of listed companies keeps rising.However,China economy is in a period of rapid development,the system is still not perfect.As a result,the incidents of financial frauds of listed companies have a high incidence.Financial fraud is worsening,more and more such incidents have occurred.Although the state has introduced corresponding regulatory measures to crack down on listed companies,the financial fraud acts also remain incessant.Some companies whitewash financial conditions for their own benefit,and provide stakeholders with financial reports which deviate from actual value.How to effectively identify financial frauds and suppress the occurrence of fraud cases is a problem that the securities market has been discussing and urgently needs to be solved.In this context,this paper analyzes and compares the effects of various improved financial fraud recognition models.This research is of great practical significance.This article is based on the non-compliance database of CSMAR.Firstly,the listed companies that were punished for financial fraud from 2007 to 2016 are picked out from the database,336 fraud firms are finalized.Based on the Beasley pairing principle,336non-fraud companies are selected as matching samples.SPSS is used to test the significance of the 29 primary indicators.Finally,factor analysis is used to reduce the indicators that passed the significant test and get 8 variables for financial fraud identification.The 8 variables are taken as the input variables.The fraud and non-fraud(non-fraud is marked as 1,and fraud is marked as 2)are used as output variables of the data mining model.The improved GA-BP neural network,decision tree and support vector machine(SVM)are used to train the training sample,then build model.The test data is used to test the model and the test effect of the model is given.The results show that the three models can effectively identify the financial fraud of listed companies,and on the basis of this model,an improved model idea is proposed: a comprehensive recognition model with superposition of multiple recognition models.Finally,combined with the results of theoretical and empirical analysis,this research puts forward some deficiencies and prospects of the future work.