Research On Financial Early Warning Of Listed Companies Based On Evidence Theory

The persistence of the new crown epidemic and the volatility of the international situation have increased the possibility of financial risks for enterprises,and the occurrence of financial risks for enterprises will increase the possibility of enterprises falling into financial difficulties,which will not only damage the normal production and operation activities of enterprises,but also lose the interests of creditors and investors,in serious cases,may also lead to the turbulence of the capital market and society.An effective financial early warning model is of great significance for the development of Chinese economy and enterprises,and it is urgent to strengthen the financial early warning of listed companies,so how to establish an effective financial early warning system has become an important issue to be solved.Existing research work on enterprise financial early warning models focuses on adjusting model parameters and changing model constraints,but most researches do not address the uncertainty in the construction process of financial early warning models,and also ignore the impact of financial data distortion caused by financial fraud on the predictive performance of financial early warning models.This paper analyzes the connotation and definition of financial distress by examining theories related to financial warning.After understanding the principles of DS-evidence theory,Support Vector Machine(SVM),Na?ve Bayes(NB),Gradient Boosting Decision Tree(GBDT),Logistic Regression(LR)and Adaptive Boosting(Ada Boost),we applied the evidence input optimization method and the evidence synthesis rules of DS-evidence theory to integrate five base classifiers,SVM,NB,GBDT,LR and Ada Boost,and programmatically implemented the construction and experiments of the evidence theory-based integrated classifier(DS-MC)financial warning model by combining the Sklearn’s three-way library and Python development platform.Considering the influence of financial data quality on the prediction performance of financial early warning models,the Benford’s factors for data quality detection was constructed using Benford’s law.The data of financial early warning index system with and without Benford’s factors were divided into two groups,and the SVM,NB,GBDT,LR and Ada Boost and DS-MC financial early warning models were used to train the two groups of data,and the output results were evaluated using Accuracy,Recall,F1 value,and Precision,and the research found that:(1)DS-MC financial early warning model solved the uncertainty problem that multiple classifiers have different classification results for the same sample,and the DS-MC financial warning model has higher prediction accuracy compared with LR,NB,SVM,GBDT and Ada Boost;(2)Benford’s law can effectively identify the fraud risk of financial data,and the Benford’s factors constructed using Benford’s law effectively improve the prediction accuracy of the DC-MC financial warning model.