Research On Accounting Information Distortion Recognition Model Of Listed Companies Based On Data Mining

In the financial reports disclosed by listed companies,accounting information accounts for more than 70% of the content.Accounting information is not only an effective basis for economic decision-making by various entities,but also a major factor for the stable development of the capital market.With the vigorous development of the company's informatization,economic transactions have gradually become networked and intelligent.In the vast amount of accounting information,the traditional methods of identifying accounting information authenticity are costly and inefficient,and can no longer meet the needs of the information age.Therefore,how to quickly extract the required data and identify the reliability and authenticity of accounting information has become a very important topic at present.Introducing data mining methods into accounting and auditing fields not only meets the needs of the public,but also has unique advantages in shortening operation time,optimizing resource allocation,reducing human subjective factors and improving recognition accuracy.In this paper,the data mining method is applied to the accounting information distortion identification of listed companies.Based on the in-depth understanding of relevant theories and the analysis of existing literatures,the data mining method is used to classify the accounting information distortion and non-distortion of listed companies.First,the initial index system is summarized through literature research,and the principal component index is formed by principal component analysis.The nine principal component indicators formed have explained the total variance of nearly 91.9%,which can largely replace the initial indicators to complete the construction of the accounting information distortion recognition model.Secondly,the financial report data disclosed by China's listed companies in 2008-2017 is selected as the research sample,and three kinds of single accounting information distortion recognition models are constructed.Support vector machine recognition model,generalized regression neural network recognition model and decision tree recognition model are constructed.Among the three recognition models,the support vector machine recognition model has the highest correct rate,reaching 94.8%.The generalized regression neural network recognition model has the shortest learning time for data,and the decision tree recognition model is easier to operate and understand.Then,the model evaluation and the construction of the comprehensive recognition model are carried out.The evaluation results show that the support vector machine recognition model has the best comprehensive performance,followed by the decision tree recognition model,and finally the generalized regression neural network recognition model.Compared with the experimental results tested by the single recognition model,the experimental results obtained by the comprehensive recognition model have higher credibility,better stability,and the recognition accuracy rate reaches 91.18%.Finally,combined with the research of this paper,management suggestions are proposed from three aspects: policy restriction,internal control,and investors and creditors' ability to identify accounting information distortion.