Research On The Identification Of Financial Fraud In Listed Companies Based On CNN-BiLSTM Model

The financial fraud of listed companies has seriously disrupted the healthy development of the capital market and harmed the interests of many investors.The antagonism betw een regulatory agencies and financial fraud companies has never stopped.How to effectively and accurately identify financial fraud companies has also become a hot spot for scholars at home and abroad.Existing studies have shown that by collecting data on the financial ratio indicators of enterprises,we can build machine learning models and deep learning models to identify financial fraud companies.However,in the existing literature,the financial ratio indicators of a single year are usually used for model construction,and the use of multi-year data is lacking.which is difficult to reflect the temporal correlation of fraud.Therefore,this paper collects and processes the multi-year financial data indicators and constructs a fusion model of CNN and BiLSTM with attention mechanism called CNNBiLSTM to use multi-year data for financial fraud identification.And this paper compares the recognition effect of the CNN-BiLSTM model with the logistic regression model and the random forest model at different year lengths.The empirical results of this article show that in the horizontal comparison of the financial fraud models,the CNN-BiLSTM model and the random forest have a better recognition effect than the logical regression model;In the longitudinal comparison of multi-year data,with the increase of the length of the year of data,the overall performance of the CNN-BiLSTM model is further improved and finally it has a better recognition effect than other models,but the recognition effect of random forest and logistic regression fluctuate or even decrease with the use of multi-year data,which reflects the advantages of CNN-BiLSTM in introducing multi-year data for financial fraud identification.This article also explores the effects of the different market sectors on the identification effect of financial fraud model recognition,and finds that there are differences in the recognition effect of models in different market sectors.The enterprises of SSE motherboard and SZSE motherboard are suitable for fitting prediction with CNN-BiLSTM model,while the small and medium-sized enterprises(SMEs)in small and medium-sized board are more suitable for fitting prediction with random forest model.At the same time,the proportion of fraud samples in SMEs is relatively high,but the identification capability of model is low in SMEs.Therefore,Institutions can strengthen supervision of the small and medium-sized board market,and auditors should be more cautious when auditing financial fraud in SMEs.