A Study On Listed Companies' Credit Risk Early-Warning Based On Machine Learning

With the continuous growth of listed companies in China,the trading scale of listed companies as the main financing body is also increasing,the credit risk contained is also increasing,and the events of credit risk are also emerging in an endless stream.In this context,strengthening the control of credit risk of listed companies and the establishment of a sound warning model will help to promote the confidence of investors in the Chinese capital market,and also help the listed companies to regulate and manage their own risks.It is of great significance to the stability and healthy development of China's financial market.First of all,the related theory of early warning of credit risk is elaborated in detail in this paper,then the index system of early warning of credit risk is designed,and the factors influencing early warning of credit risk are analyzed,and the financial and non-financial indexes are selected.Subsequently,the selected data are used to construct the early warning model of credit risk of listed companies,and combine the empirical results to compare three kinds of early warning models based on machine learning.Finally,we make a brief summary and forecast on the work of this paper.In the process of establishing the early warning model of credit risk,the data preprocessing is conducted to the selected financial and non-financial index,the principal component analysis in factor analysis is used to reduce the dimension of early warning index,and the model is established by using three kinds of data mining algorithms including neural network,support vector machine and random forest.By comparing the prediction accuracy of the model separately,the random forest model is determined to be the relatively optimal early warning model.Later,on the basis of Gini index in random forest model,we get the contribution of each input variable to the random forest early warning model,and combine with the relevant index represented by the relevant factors to carry on the realistic analysis.Finally,it is the model inquiry simulation.the relationship between the credit risk of the listed company and the performance of the listed company's stock in the secondary market is explored.A further analysis is conducted in combination with the results.The empirical results of credit risk early warning of listed companies based on different machine learning algorithms show that the random forest early warning model is more suitable for the early warning of credit risk of Chinese listed companies,and its prediction accuracy is as high as 98.53%,and the stability of the model is quite high.(Standard deviation is only 0.01).More importantly,the model can maintain the high accuracy and stability of the machine learning model,and at the same time establish the relationship between the index variables and the prediction results to a certain extent,breaking the shortcomings of most black-box models.So,people can have a more specific and intuitive understanding of the early warning results.Finally,in the established random forest warning model,the credit risk of the listed company and the performance of the company's stock in the secondary market are explored.The research shows that the credit risk of the listed company is not completely consistent with the performance of the company's stock in the secondary market.Corporate credit risk can only explain the performance of stocks in the secondary market to a certain extent.