Research On Credit Risk Of SMEs Based On Deep Learning

At present,China attaches great importance to the development of small and mediumsized enterprises(SMEs).The Central Economic Work Conference and the 14th Five-Year Plan have both made important arrangements to serve the innovative development of SMEs.However,SMEs have financing problems because of the high information asymmetry between enterprises and financial institutions.At present,the assessment of SMEs credit risk mostly focuses on traditional quantitative data such as financial statements,but in fact,such traditional credit research based on quantitative data cannot target the characteristics of SMEs.The information asymmetry between enterprises and financial institutions cannot be fundamentally solved.From this point of view,it is very important to study credit risk according to the characteristics of SMEs.First of all,based on the advantages of massive information brought by big data technology,text mining technology is used to dig the relevant data of SMEs.In terms of quantitative financial reporting data,140 financial index pools of four evaluation dimensions were established,and the most valuable 20 financial characteristics were obtained by feature screening method.In terms of qualitative enterprise characteristic data,a total of 18 qualitative indicators were obtained from four evaluation dimensions.At the same time,the text mining technology is used to further analyze 5 unstructured text data indicators and get 21 three-level indicators.The experimental results show that both the quantitative and qualitative indicators proposed in this study are valuable,and the qualitative enterprise characteristic indicators play an important role in complementing the traditional financial indicators.Secondly,on the basis of the SMEs credit risk index system,a credit risk assessment framework for SMEs is developed,which integrates multiple credit risk influencing factors,time span and classifier models,and provides insight into the credit risk of SMEs.The experimental verification confirms that the deep learning model shows certain advantages compared with the traditional machine learning model in predicting the credit risk of SMEs.Finally,the Attention mechanism is innovatively used in the deep learning model,and the LSTM credit risk measurement model based on the Attention mechanism is constructed.Compared with the general RNN and LSTM deep learning models,it is found that the Attention mechanism based deep learning model has a better performance in predicting the credit risk of SMEs.