Research And Implementation Of Intelligent Financial Statement Fraud Detection System

Financial statements are the basic documents that reflect the financial situation of enterprises.The occurrence of fraud will make the financial statement data lose its own authenticity.It will not only mislead the users of financial statements,lead to their loss of confidence and enthusiasm in the securities market,hinder corporate managers from making correct management decisions,expand the operational and financial risks of enterprises,but also attack the normal order of the capital market.With the development of the information industry economy,the business operation mode of the real economy is becoming more and more complex and the mode of operation is becoming increasingly diversified.At the same time,the means of financial statement fraud are constantly updated,which makes it difficult to identify false financial statements and has brought great challenges to financial audit.Therefore,it is of great practical significance to establish an intelligent financial statement fraud detection system with data mining technology to assist auditors and relevant supervision departments to improve the level of fraud identification.Using the data of financial statements and governance structure of A-share listed companies in China from 2007 to 2018,this paper process a random forest fraud detection model based on SMOTE oversampling and simulated annealing algorithm feature selection.Then an intelligent fraud detection system is constructed based on this model.The main work of this paper is as follows:1.This paper makes a comparative study between different feature selection methods and data mining classification models.57 financial and non-financial indicators are selected comprehensively through literature research.Relief method,genetic algorithm and simulated annealing algorithm are introduced to screen the original indicators based on four classification models which are logistic regression,BP neural network,support vector machine and random forest.Different combinations of feature selection methods and classification models are evaluated by appropriate evaluation methods.The evaluation results show that genetic algorithm and simulated annealing algorithm perform better than Relief,and the combined model of simulated annealing algorithm and random forest has the best performance.2.Taking into account that the random forest models have high precison but low recall,SMOTE algorithm is used in this paper to eliminate the influence of class imbalance on the identification effect of fraud samples through oversampling the training data set.This method increases the recall rate of the combination of simulated annealing algorithm and random forest model on the test dataset up to 90% and further improves the overall performance of the model.3.Based on the model selected in the experimental stage,an intelligent fraud detection system is analyzed,designed and implemented.With providing basic user management and data management functions,the system can assist users to establish and use fraud detection models.It can be used as an effective auxiliary means to make up for the deficiencies of manual audit and improve the efficiency of audit work of auditors and supervision departments.