Research On The Construction Of Financial Fraud Identification Model Of Listed Companies Based On Deep Learning

In the past several years,owing to the swift progress of the capital market and the economy,the number of listed companies has increased significantly,and many of them have also experienced substantial growth in size and scale.At the same time,the accompanying financial fraud problems also appear from occasional to common,the circumstances are more and more bad,and the consequences are more and more serious.From the data collation starting in 2017,it can be seen that the administrative punishment notice issued by the CSRC involves more and more penalties for financial fraud,and more and more listed companies involved have been identified as financial fraud companies.Through sorting and analysis,it can be seen that the problem of financial fraud identification exists widely and has not been well solved.In this paper,SMOTE sampling data is used to train DT,RF,ET,XGBoost,KNN,LR,LGBM and Ada Boost to build single model scoring and fusion model effect,and select the best three algorithms RF,ET and XGBoost to find the characteristics that have a great influence on financial data fraud in publicly listed companies.In the case of optimal AUC index,the feature selection results of the three algorithm models of RF,ET and XGB are integrated.The feature importance weight values were computed,and the top 31 features were selected as financial indicators of listed companies based on their relevance to fraud characteristics.For financial data of listed companies since 2015 and CSRC punishment documents,5125 A-share listed companies were selected as the data set,including 3336 companies in manufacturing industry and 1789 companies in other industries,2730 companies in 3336 manufacturing enterprises in the data set as training sets,the remaining 606 as test sets;1464 enterprises in 1789 other industries as training sets,and the remaining 325 as test sets.Finally,this paper selects ST Kangmei,a representative company in the leading pharmaceutical manufacturing industry,as the research subject.Its disclosed a said nearly three hundred billion “accounting error” correction announcement,this is a billions “white” financial fraud,belongs to the malicious use of early accounting error correction,and this means to hide the financial fraud the fact,this is the management intentionally and illegal operation,shock the capital market,financial fraud problem is not only widespread in regulation may have omissions of small and medium-sized enterprises,is also common in large enterprises.ST Kangmei was replaced into the fusion model for validation and the model was successfully identified ST Kang mei fraud year.With the same method,we tested the three enterprises of Jinya Technology,Erkang Pharmaceutical and Lianjian Optoelectronics,and all successfully identified the fraud year of each enterprise,and the model verification was effective.Through establishing the model in this paper,we also know that the choice of influence factor is very critical for the model establishment.