The Neighborhood Rough Set And Support Vector Machine Combined Model For SME Credit Assessment

From the process point of view of world economic development, small andmedium sized-enterprises (SMEs) has been the driving force of economicdevelopment of countries. In China, the number of registered SMEs has reached10,427,400in2009. It makes up more than60%of China’s GDP. The problem ofdifficult in financing, which caused from the information asymmetry between SMEsand banks, of SMEs, however, is still main factor to restricting its development.Therefore, reasonable assessments of the credit situation of SMEs play key role insolve this problem.Through analysis of the characteristics of SMEs and credit status, combined withexisting enterprise credit assessment methods, this thesis comparative analyses theapplicability of various types of credit assessment method for SMEs. We believe thatexpert scoring subjectivity is too strong. Experts in the assessment process will leadto selection bias because of their own knowledge structure deviation; Traditionallinear credit scoring models has strict requirement of sample data. It performed poorlyin the SMEs credit assessment, caused SMEs‘data relative missing; KMV modelneeds lots of effective market information. It also performed poorly in the SMEscredit assessment, caused time to market for SMEs is too short. Credit Risk+model aswell as performed poorly in the SMEs credit assessment, caused the related parties ofSMEs are too much. On the contrary, machine learning methods can effectively applyto SMEs assessment, because of less demanding on the sample data and self-learningability. Then, this paper discussed NRS and SVM principle of operation and thespecific algorithm and compared the advantages and disadvantages of both methods.Then, we build a NRS-SVM model. Firstly, in the process of the establishment ofSME credit evaluation index system, this article introduced neighborhood rough settheory to reduce indicators. And then we set the simplified index system as the inputof neutral network to make categorical forecasting.At last, this paper use the2008to2009, SME Board‘s data to empirical study theNRS-SVM model. The conclusion is that compared with only using SVM, theNRS-SVM can effectively remove the redundancy index; compared with traditionalcredit scoring models, the NRS-SVM can better adapt to the SME credit assessment.