A Research On Financial Fraud Identification Model Of Listed Companies Under Data Mining Technology

Financial fraud cases have occurred in successional worldwide,from early years Enron in the United States,Parmalet in Italy,Olympus in Japan and YinGuangXia in China,etc.,to China's Green Land,Nine Top Group,WanFu Biotechnology and KangMei Pharmaceutical Co.,Ltd.in recent years.Financial fraud cases have emerged repeatedly at home and abroad,and the means of fraud have become more diverse.According to the Association of Certified Fraud Examiners?ACFE?,the number of financial fraud cases examined in 2018 increased by11.6%over 2016,and the amount of damage involved was as high as$7 billion^[1].Although China's capital market has not been established for a long time,there are also many cases of financial fraud cases.The behavior of listed companies being financial fraud not only affects their development,but also influences the interests of creditors and investors,hinders the steady development of China's capital market economy,and seriously affects the social economy.Therefore,the problem of financial fraud has always been a challenging and hot topic in academic research.This paper applies data mining technology to the research of financial fraud identification of listed companies.After combing the existing relevant literature,the annual report data of China's Shanghai and Shenzhen A-share listed companies from 2009 to 2018 is used as the research object.Fraud samples were selected from listed companies that were punished by China Securities Regulatory Commission,Shanghai and Shenzhen Stock Exchanges,and non-fraud samples were selected at a 1:1 ratio according to certain criteria.After screening,the data from 684 companies were selected as research samples.About selecting variables,this article is based on the Theory of fraud risk factors.Based on the financial fraud methods commonly used by listed companies and the results of existing literature research,the variable indicator system of this paper is determined.A single financial fraud recognition model was build by using Support Vector Machine,BP neural network and Random Forest which are three data mining technologies.And then a better recognition of financial fraud comprehensive recognition model will be created by combining these three data mining techniques.Through research,the basic conclusions of this article are:?1?Among the single financial fraud recognition model,the financial fraud recognition model that based on Support Vector Machine has shown the best recognition effect,followed by which were based on the BP neural network and the Random Forest.The overall recognition accuracy of the three single models is between 69%and 75%.?2?In the comprehensive identification model of financial fraud,this paper builds a comprehensive identification model of financial fraud by combining three data mining technologies:Support Vector Machine,BP neural network and Random Forest by giving each single identification model different weights.After evaluating the model,it was found that the comprehensive identification model of financial fraud under data mining technology has better identification effect than the single identification model.The contribution of this paper lies in the application of data mining technology in constructing of financial fraud identification models,the construction of financial fraud identification models based on three types of data mining techniques:Support Vector Machine,BP neural network,and Random Forest.A comprehensive fraud recognition model was established that based on a single fraud recognition model with 78%accurate rate,an80%precision rate,77%recall rate and an 78%F₁ score.It is concluded that the comprehensive identification model of financial fraud under data mining technology has better identification effect than the single identification model.