Research On Credit Risk Prediction Of Real Estate Enterprises Based On Deep Learning

As China’s economy continues to explore the path of high-quality development,the issue of corporate credit risk has become increasingly prominent.Nowadays,it has become an urgent task to predict the credit risk status of enterprises and to provide timely warning.This study aims to combine enterprise business information with machine learning and introduce deep learning algorithms to predict enterprise credit risk.Through this effective prediction model,we can grasp the actual business situation of enterprises and predict potential credit risk problems,and strongly address the challenge of credit risk.At the same time,the contribution of this study is to improve the ability of the state and society to predict corporate credit risk and provide some reference basis for guiding the business operation of enterprises.The importance of this study lies not only in its application to corporate credit risk prediction,but also in its ability to provide important ideas and methods for the practical application of deep learning in the field of corporate credit risk,thus injecting new vitality and vigour into the relevant research field.This dissertation is an analytical study of various types of business information of listed real estate enterprises.The dissertation is divided into three parts: First,we construct a BERT-Bi LSTM-ATT model and introduce text sentiment evaluation indicators.The text content of "Management’s Discussion and Analysis" in the annual report is annotated with sentiment features,pre-trained with the BERT model,and then double semantic extraction of text information through the Bi LSTM model to finally obtain the sentiment score index.In the process of model construction,different from the previous traditional models,the output of the last four coding layers of the BERT model was combined through experiments and finally input into the Bi LSTM model as input variables,which improved the performance of the combined model.Secondly,the XGBoost model was used to screen enterprise credit risk characteristics indicators and construct an evaluation index system.As there is an imbalance in the enterprise credit risk data,the SMOTE algorithm is used to expand the data set to achieve the purpose of data widening and provide data support for subsequent model training.The XGBoost model was used to discern the degree of correlation among the credit risk indicators and output the importance ranking of each feature to determine the indicator factors affecting the credit risk of enterprises.Thirdly,a study was conducted to predict whether an enterprise is risky by constructing an improved CNN-ILSTM prediction model.The CNN model is fused with the LSTM model and the residual concatenation technique is used to overcome the representation bottleneck problem of the LSTM model,so as to output the enterprise credit risk prediction results.The results of the study show that the analysis of the text of the Management Discussion and Analysis(MD&A)reflects the future development tendencies of the company and extends the system of indicators for the evaluation of corporate credit risk.The model solves the problem of data imbalance and identifies factors influencing credit risk.The final prediction of whether a company is a credit risk is 97.4% accurate.Through comparative analysis with other single or combined prediction models,it is found that the prediction model constructed in this study demonstrates higher prediction accuracy and stability,enriches the prediction method of enterprise credit risk,and provides suggestions on the prevention of enterprise credit risk from the management perspective,providing reference for the actual operation of enterprises.