Research On Detecting Financial Statement Fraud Based On Weighted Borderline-SMOTE-RF

With the vigorous development of China's information technology industry and increasingly fierce market competition,in the increasingly fierce market competition environment,in order to meet the needs of maximizing its own interests,companies in this industry have become high-frequency areas where financial statement fraud occurs.Constructing an effective and complete financial fraud identification framework for listed companies is conducive to helping investors make decisions.At the same time,it helps regulators to make accurate judgments and efficiently perform massive audits,thereby promoting the rational operation of the entire securities market.Reviewing the relevant research of scholars at home and abroad,the problem of detecting financial statement fraud has been rich in results for decades.However,there are few existing domestic and foreign literatures to select the information technology industry as the research object.The Internet technology industry is different from traditional industrial financial reporting information.The financial indicators of this industry have their own characteristics.This paper hopes to provide reference information for the audit of information technology industry enterprises in feature extraction.Previous studies in this field were mostly based on the principle of equal pairing for sample processing.Considering the practical significance of the cost of misclassification in financial fraud identification,this paper combines the unbalanced classification problem to improve the overall recognition accuracy and the identification accuracy of fraud samples.Based on the current market development trend,this paper focuses on the A-share information technology industry.According to the financial characteristics of the industry,it simulates the unbalanced classification problem in the real audit environment and constructs a financial fraud feature extraction,sample synthesis and financial statement fraud identification.The complete and effective framework.The innovation and research work of this paper are as follows:Firstly,this paper constructs a new framework for identifying financial fraud.This paper selects 585 Chinese listed A-share information technology enterprises as research samples from 2007 to 2017.Empirical research shows that the overall accuracy of this framework is 89%,and it can effectively identify financial statements fraud.Secondly,this paper innovatively proposes a cost-sensitive Relief feature extraction sub-model,which extracts eight features from 38 original feature indicators as the core feature subset,so that the total misclassification cost of the classifier is the lowest.Thirdly,combined with the optimal feature subset,this paper proposes a sample synthesis sub-model based on weight Borderline-SMOTE,which synthesizes samples from multiple feature dimensions to improve the authenticity of samples and solve the problem of unbalanced classification of samples.Fourthly,the performance results of different sample synthesis algorithms and different recognition models are analyzed by empirical comparison.The results show that the framework proposed in this paper has better recognition effect.Based on this paper,a complete framework for financial report fraud identification is constructed,which provides relevant suggestions for the company's own development and regulatory agencies: On the one hand,information technology enterprises themselves need to develop sound financial management system according to the characteristics of the industry to enhance the awareness of risk and cost management,on the other hand,they need to form a sound internal control mechanism to monitor and control from the core financial indicators.Actively carry out risk management,identify problems and timely strategic adjustment;regulators need to break the traditional industry consolidation system,establish a reasonable and flexible assessment system and comprehensive financial indicators supervision system.