Recognition Of Financial Fraud Of Listed Companies Based On Data Mining Methods

The listed companies in our country have been underwent development indecades,the securities market has grown increasingly that promote the nationaleconomic growth effectively,as well as complete the market economicsystem.However,the financial frauds of listed companies took place repeatedly,thoseevents were not only lead to a large amount of loses for companies themselves,butalso suffer the interests of companies’ stakeholders,and severely impair thedevelopment of securities market.All the time,the issues about detecting financial frauds are the hot and difficultquestions in relevant studies.In early studies,according to the companies’ publishedfinancial data,most researchers used the common statistical regression methods todetect the frauds,and obtained certain results to warn and detect the frauds.However,inrecent years,the financial fraud issues have been increasingly difficult to warning,thereasons are that companies’ fraud methods have become more covert,it hard to findthe evidences of misrepresenting the financial data.Therefore,to detect the financialfrauds no longer used the companies’ financial data,but should base on the externalenvironment indices that the company is hard to control,used the artificialintelligencemethod,such as data mining,to confront the covert frauds.This paper based on above idea.First,by summarizing and settling the relatedliteratures in domestic and overseas, the theory and methods of financial fraudidentification, as well as their merit and demerit were analyzed. In theoretical side,this paper were summarized and defined the financial fraud, corporate environmentand data mining technology, then analyzed the influence of environment on financialfraud. Second, the paper presented a relatively complete evaluation method, toevaluate the degree of listed companies’ frauds, and then distinguished the fraudulentcompanies into serious fraudulent companies and common fraudulent companies. Onthis basis, the empirical research was screened proper samples and internal andexternal environment indices, used the Support Vector Machine (SVM), Bayesian Network (BN) and Decision Tree (CART) to detect the conducts of listed companies’financial fraud, then built the model of detecting financial fraud. Empirical resultsshowed that that three data mining methods had the average detection rates more than70%, SVM and CART had a relatively stable detection results. When integrated thatthree independent classifiers, the detection rates could been reach even more ideal.Finally, the model was tested, and received the final conclusion, to provide thenecessary theoretical and methodological supports for future research of detectingfinancial frauds.