Research On Machine Learning Based Chinese Company Financial Risk Detect System

The risks in financial markets are ubiquitous,and the operation of listed companies is affected by various factors.Therefore,the research on the financial risks of listed companies is also of great significance.Traditional financial statement-based research methods such as the five-factor Z-score model have great significance in the field of financial risk prediction,but they still have their limitations.On the one hand,most of the traditional statistical analysis models use only a small number of financial indicators.In addition,most of the traditional models for textual information have no ability to analyze.With the advent of the era of big data,the above two limitations can be partially solved.In recent years,with the continuous development of computer technology,Fintech has gradually become one of the main research directions in the financial field.Machine learning,artificial intelligence,and deep neural networks are gradually being recognized and understood by more people.They have left a deep impression on image recognition,automatic driving,and natural language processing,and prove that it has unlimited potential.The power of deep learning lies in the ability to build complex model structures and a large number of parameters,based on which to extensively mine and learn massive amounts of historical data.The financial sector is a data-filled environment.Companies with financial risks often find clues in their financial status,operational status,and public opinion before the crisis erupts.Therefore,how to construct a set of financial risk identification system based on machine learning for listed companies,and effectively mine and utilize a large amount of data,so as to achieve efficient supervision of listed companies is a problem worthy of further studyThis article uses the report data of the listed company and the annual report text data separately.For the financial statements of listed companies,we use the decision tree,random forest and XGBoost model to mine and compare the classification effect.For the text data of the listed company's annual report,after cleaning,the paper uses the pre-training word vector to digitize the text,constructs a classification model based on deep neural network,and optimizes the model using the improved loss function.Finally,the results of the two models are combined to obtain the final predicted value of the company's financial risk probability.Our empirical results show that the comprehensive model based on random forest and XGBoost has achieved better classification results,and the addition of text information can slightly improve the classification efficiency of the model.Among them,the XGBoost model reached the highest comprehensive score,while the random forest model,although the comprehensive score was slightly lower than XGBoost,achieved a recall rate of 82.53%,that is,82.53%of the companies with financial risks in the test set were modeled.Successfully identified.Since the loss of a risky company in the financial market far exceeds the misjudgment of a normal company,the recall rate is more important than the accuracy rate in this scenario.Therefore,the model with the best result is based on this paper.A model of a random forest.