Research On Financial Fraud Detection Method For Listed Companies Based On Fusion Model

As market competition intensifies and investor demand rises,financial fraud by listed companies is receiving more and more attention.In the face of increasingly complex market environment and financial data,traditional financial analysis methods can no longer meet the actual needs.It is of great practical importance to improve the accuracy and reliability of financial analysis and to provide investors and regulators with more effective risk warning and supervision tools.At present,although many researchers have conducted in-depth research on financial fraud detection techniques for listed companies and proposed various detection methods,there is still some room for improvement in the detection accuracy of these methods.For this reason,further in-depth exploration of financial fraud detection methods is needed to improve the accuracy and reliability of detection.This article takes the financial data of listed companies as the research object,firstly,pre-processes the data,feature selection and data imbalance processing,then designs the classifier model and the corresponding detection algorithm,and finally validates the design method through experiments,and achieves a better detection effect,the specific research content is as follows:(1)To address the problem of unbalanced financial data of listed companies,this article improves the SMOTE sampling algorithm on the theoretical basis of studying unbalanced financial data sampling algorithms,designs the ISMOTE algorithm,and combines the penalty term-based and tree-based feature selection methods to design a method for detecting financial fraud of listed companies based on the improved SMOTE algorithm and random forest.The experimental results show that the method exhibits high accuracy on the Teddy Cup dataset and effectively addresses the negative impact of data imbalance on the model effect.(2)To address the limitations of single data source and the problem of single model generalization error,this paper adds unstructured textual information to the Guotaian dataset to form a multi-source heterogeneous dataset.Based on ISMOTE sampling algorithm,we combine Transformer to select financial data features,and fuse ant colony algorithm to improve random forest,and finally use Stacking model fusion technique to design a fusion model-based method to detect financial fraud of listed companies with multiple data sources.After experiments,the fusion model outperforms the single model such as Random Forest in detecting financial falsifications in the Guotaian dataset with better accuracy and no overfitting phenomenon.(3)Based on the above financial fraud detection method,a system capable of detecting financial fraud in listed companies has been developed.The system not only enables the management of financial data sets,data pre-processing,feature selection and imbalance data processing,but also allows the training of classifier models and provides a display of the final falsification detection results.