Credit Evaluation Model And Application Based On Random Forest

With the continuous improvement of China's market economy system,credit evaluation has gradually become an important foundation of social and economic development,and the bond market has also become the cornerstone to improve our country's financial market.Therefore,studying the credit problems of the bond and constructing credit evaluation model have played a great role in promoting the vigorous development of our country's credit business,and become an inevitable choice of the development of market economy.This paper takes the bond of our country's listed bond companies in 2017 as empirical research objects,establishes a set indicator system of bond credit evaluation by constructing random forest model,and calculates the weights of the credit evaluation indicators based on the importance of the indicators,and then constructs the fuzzy integral credit evaluation equation and divides credit ranks on the basis of credit evaluation results.The main conclusions of the study are as follows:(1)On the basis of the selection index set for the credit evaluation of bond,constructs and optimizes the random forest model,and calculates the importance values of the indicators according to the modeling results.Based on this,the credit evaluation indicators with significant discrimination ability for default status are retained,and finally a set of credit evaluation indicator system including 13 indicators is constructed.At the same time,comparing the indicator system with the international "5C" evaluation principle,the results show that the construction of the credit evaluation indicator system is reasonable.Furthermore,the ROC curve determines that the indicator system has a strong ability to predicate the default status and so its effectiveness is high.(2)Confirm the weight coefficients of the indicators according to the importance values,and obtain the weight values of the corresponding criterion layers.The data show that the external macro-environment is the main factor that affects the credit degree of the bond,followed by the financial factors and non-financial factors.These factors are closely linked and have an important influence on the credit evaluation of the bond.(3)Construct the fuzzy integral equation to calculate the credit scores and divide the credit rating of the evaluation samples.The results show that the better the credit status of the bond,the higher the credit rating,the higher credit rating contains fewer default samples,the overall credit rating distribution is roughly bell-shaped,and the credit level of most bond is grade A,BBB and BB,which conform to the actual credit level of the current bond market.The innovations of the research lie in that firstly the importance values of indicators are measured through random forest optimal model,and the credit evaluation indicator system that has an important influence on the credit evaluation results is screened,which reflects a new idea of screening credit evaluation indicators under the big data non-parametric environment.Secondly,use the importance of indicators to measure the weights of indicators,reflecting the idea of weight measurement that the stronger ability to identify default status,the greater the weight of indicators.The characteristics of the study are as follows:firstly,to improve the imbalance phenomenon between default samples and non-default samples in real credit data,and to enhance the classification accuracy of default samples and non-default samples.Secondly,to use fuzzy integral method to establish the credit evaluation equation,which overcomes the disadvantages of the traditional classical linear weighted evaluation model.