Multi-classification Early Warning Research On The Financial Status Of China's Internet Finance Listed Companies

In recent years,Internet finance,a new form of financial services,has developed rapidly.With the support of Internet information technology,P2 P network lending,mobile payment,crowdfunding,and information intermediary services have greatly improved the efficiency and quality of financial services.The development of inclusive finance has deepened the reform of the financial system.At the same time,as a new thing,the rapid development of Internet finance is accompanied by many problems.Through the initial “barbaric growth”,various potential risks have become increasingly prominent.Many Internet financial companies are not well managed,and various financial crises have emerged.Investors have caused serious losses,which is not conducive to the smooth operation of the entire financial system.It is precisely because the development of the Internet financial industry involves multiple interests,it is necessary to construct a financial early warning model for China's Internet finance listed companies.This paper first expounds the definition of financial crisis and financial early warning theory,and summarizes the research results of scholars in financial early warning.It is found that in the past,when financial warnings were issued to listed companies,the financial status was always judged according to whether the sample data was marked as ST.Such division was obviously not detailed enough to reflect the real financial status of each enterprise,so consider more The method of classification.Combined with the current development status of China's Internet finance industry,208 sets of sample data from 2016 to 2017 were selected as research objects,and the financial early warning indicator system of China's Internet finance listed companies was constructed.Through cluster analysis,the financial status is classified and evaluated,and the validity of the clustering results is tested.The financial status of the research sample is divided into three levels: “health”,“general” and “poor”,so that they are different.A degree of financial warning,in which a financially poor company is more likely to have a financial crisis.Then use SPSS23 statistical software to analyze the indicators to remove the influence of multicollinearity,and use seven principal components to represent all early warning indicators.Finally,the sample data is divided into training set and test set in a 7:3 manner,and a financial early warning model based on multi-class SVM and multi-logistic regression is established.For the SVM model,in the selection of kernel function and kernel parameters,the advanced research results of machine learning at home and abroad are used,and the model is trained and adjusted by e1071 in the R software package,and the grid point search method is used to determine the optimal parameters.The optimal prediction model is obtained,and the prediction effect of the model is analyzed and compared by using the confusion matrix and AUC values.The empirical analysis shows that combining unsupervised learning with supervised learning algorithms overcomes the bipolar state of ST and non-ST in the past financial early warning research,and can conduct more detailed early warning research on corporate financial crisis.At the same time,the multi-class SVM model is superior to the traditional multi-logistic regression model when dealing with small samples.Therefore,in the future processing of unlabeled sample data sets,K-means clustering and SVM model can be used to conduct multi-level financial early warning research on China's Internet financial listed companies.