Research On Financial Fraud Identification Of Listed Companies

The growth rate of the number of listed companies is gradually increasing.As of October 2021,there are5243 listed companies in China,including 4512 A-share listed companies.However,since the capital market has been built,the companies’ financial frauds are always being.This behavior maliciously destroys the market order.Due to the concealment and complexity of financial fraud,the fraud is not found in time,and it is often found out and punished after several years,with low illegal cost.The neural network model has the advantages of high-performance processing method under massive data,deep learning ability and associative storage ability.Therefore,this paper focus on the recognition effect of traditional LR model,find the NN model parameters with the best recognition effect.This paper studies and analyzes the theory and practice of financial fraud and fraud identification,compares the characteristics and shortcomings of current fraud identification methods,and analyzes the advantages of neural network model.In the empirical part,firstly,the index system is constructed and samples are selected.This paper selects 1687 fraud and non fraud samples of listed companies from 1999 to 2020 according to the ratio of 1:2;Then the logistic regression model,radial basis function neural network model and multi-layer perceptron neural network model are constructed to test the samples,and the multi-layer perceptron model with good recognition ability is selected for parameter optimization.The innovation of this paper is to adopt a new research perspective and get new research conclusions.By comparing the performance of machine learning and big data mining technology in identifying financial fraud,it not only comes to the conclusion that the neural network model has a good identification effect,but also optimizes parameters to improve the identification effect.Research conclusions:(1)the initial recognition accuracy of radial basis function and multilayer perceptron model is 77.7% and 81.9%respectively,which is much higher than that of logistic regression model(66.63%);(2)When the multi-layer perceptron neural network model is selected and the number of neurons in the first and second hidden layers is 13 and 10 respectively,the recognition effect of financial fraud is the best,and the accuracy can reach 84.77%.To sum up,this paper provides a reference for optimizing the neural network technology applied to fraud identification,lays a foundation for the research of financial fraud identification model,and provides ideas for investors to understand the fraud of listed companies.