Research On Financial Crisis Early Warning Of Listed Companies Based On Integrated Algorithm

With the economic growth and the improvement of financial supervision,more and more enterprises are raising capital through coming into the market.In the recent 20 years,the stock market in our country has developed rapidly.At the same time,with the crisis of listed companies intensifying,the issue of early warning of financial crisis has also attracted the attention of more and more experts and scholars.The use of statistical methods,machine learning and data mining technology to build early warning model of financial crisis early warning of listed companies' financial risks is conducive to discover in time and avoid risks for listed companies,and also conducive to rational choice and timely stop-loss for investors.In this paper,we select 47 listed A-share listed companies that have taken the delisting risk warning as the financial crisis samples in 2017,and 141 financially healthy A-share listed companies randomly as the matched samples at a 1: 3 ratio.In the selection of index,this article include two aspects of financial index and non-financial index,Through the initial selection,we select eight categories of a total of 60 indicators.This paper overcome the blindness to use T test or the subjective selection when selecting the index in many studies.And using neighborhood rough set theory instead of the classical rough to reduce the index,streamline the index and select the most important index.Next,This paper first uses the xgboost model and the random forest and xgboost fusion model for the financial crisis early warning field.the financial crisis warning model based on random forest and xgboost is established respectively with the indexes after reduction,and the forecasting result of xgboost and random forest model is fused by simple weighted voting to get the final model of financial crisis early warning.In order to prove the validity of the neighborhood rough sets,this paper also established a classical rough set-random forest,the classical rough set-xgboost model,the results show that the use of neighborhood rough sets instead of the classical rough set for index reduction get a better result.During the model phase,this paper uses SVM,ANN and LR classification methods to compare and analyze.The empirical results show that xgboost and random forest models are superior to SVM,ANN and LR models in all aspects.Finally,random forest and xgboost models is fused by a simple weighted voting to get an optimal financial crisis early warning model,It further confirms the applicability in the study of the financial crisis early warning.