A Multi-criteria Decision-making Method For Ordered Classification Of Small Samples

With the rapid development of small and medium-sized enterprises,small and medium-sized enterprises have gradually become an important part of China's economic development,which has a positive effect on the stability of the banking industry and finance,and even the further development of social productivity.However,the current level of internal credit management and the marketization degree of external credit services in China's small and medium-sized enterprises are relatively low,leading to lack of credit.This is the main reason for the difficulty of financing for small and medium-sized enterprises.To solve this problem,it is urgent to establish an effective credit risk assessment model.However,due to the small amount of data in small and medium-sized enterprises and the difficulty in obtaining credit data,how to accurately assess the credit risk of small and medium-sized enterprises has always been a major problem faced by academics and industries at home and abroad,and no unified and accepted method has yet been formed.Starting from considering credit risk assessment as a classification problem,researchers and credit rating agencies mostly use some statistical and data mining methods to develop credit rating models.This paper explores the potential of a small sub-ordered multi-criteria decision making method(ELECTRE TRI method)from the perspective of operations research management.Existing research shows that multi-criteria decision-making technology is very suitable for building credit risk assessment models.It enables analysts to introduce their preferences during the model construction process and understand the characteristics of the rating model.Based on the ELECTRE TRI method,this paper attempts to combine the optimization tools of two model fitting processes,that is,mixed integer linear programming and differential evolution algorithm,to build a credit evaluation model for small and medium-sized enterprises,and conduct empirical analysis on 97 small and medium-sized enterprises in China.And compared with the logistic regression model and the KNN model,comprehensively evaluate the advantages anddisadvantages of the credit risk assessment model constructed in this paper.In this regard,this article mainly does the following research.First,in the framework of the ELECTRE TRI method,a theoretical model based on mixed integer linear programming is derived by simplifying some parameters of the model.Second,considering that there may be no solution to the linear programming model,and in order to incorporate all the parameters of the ELECTRE TRI method into the model,this paper attempts to establish the ELECTRE TRI model framework using the differential evolution algorithm.Finally,the above theoretical model is empirically analyzed using the actual data of small and medium-sized enterprises in China,and compared with logistic regression models and KNN classification algorithm.The research results show that the ELECTRE TRI model proposed in this paper is very suitable for credit rating,because first of all,the model provides good classification results,whether it is evaluated from the prediction accuracy,or the performance of the classification model itself(AUC value),the overall performance is better than the logistic regression model and the KNN classification algorithm.Second,from the perspective of the interpretation effect and analysis ability of the model,our model can effectively gain insight into the relative importance of the evaluation criteria,and at the same time,it can provide credit rating analysts with more information on the role of decision attributes and corporate characteristics.This is reflected in: first,the importance ranking and selection of the ELECTRE TRI model are consistent with the results of the logistic regression model;second,compared with the logistic regression model,only a fixed coefficient value can be obtained,the parameter setting of the ELECTRE TRI model is flexible and diverse,including various aspects such as consistency test and rejection intensity.In addition,under the DE algorithm model,the non-compensatory advantages of the ELECTRE TRI model can also be well utilized.