Research On The Application Of Data Mining In The Recognition Of Financial Fraud In Listed Companies

In recent years,the number of financial fraud cases in the country has been frequent,and all kinds of financial frauds have been banned for many times.This has brought incalculable economic losses to investors,and has also brought serious negative impact on the healthy development of the capital market.How to quickly and accurately identify financial issues?Fraudulent companies have attracted a great deal of attention from investors,audit institutions and government agencies.This article refers to Kirkos et al.'s research on the use of data mining techniques such as neural networks and decision trees based on financial data of Greek listed companies to identify counterfeit behaviors,and the comparison of recognition performance of different data mining technologies by Ophir Gottlieb et al.based on financial data of US listed companies.Using the database of financial indicators of listed companies in China,research and analysis of data mining and identification technologies for the financial fraud of domestic listed companies.In this paper,a lot of research has been done on the theory and status quo of financial report fraud.Based on the theory of fraud risk factor,this paper selects from the aspects of stress factor,opportunity factor and the possibility of financial fraud discovery as the starting point for the selection of feature variables.48 financial and non-financial indicators participated in the empirical analysis.Then,by counting the year distribution of financial fraud data in the capital market in recent years and the distribution of industries,we selected 2010-2016 as a timeline with high incidence of frauds,and three major financial falsified areas-manufacturing,wholesale and retail,and real estate.The annual report data of listed companies in the industry are used as empirical samples for empirical research.The use of neural network models(BP neural network,RBF neural network)and decision trees(C&AT,QUEST,CHAID,and C5.0 algorithms)were used to identify the fraud recognition of financial reports in three industries.The empirical research results show that the recognition effect of BP neural network is better in the two neural network models.The decision tree algorithm in C5.0 has outstanding performance in the decision tree model.Overall,the applicability of C5.0 algorithm and the recognition accuracy rate are better than others.In addition,when judging whether the listed company's annual report is fraudulent,several indicators that the three industries need to focus on are:whether there are defects in internal control,ratio of retained assets to assets,total operating cost rate,working capital(capital)turnover rate,and auditors' audit opinion.