Corporate Credit Rating Based On Neural Networks And Expert Systems Research

Commercial banks must be cautious to whether they should award credit to corporations and how much they can offer to them,according to their credit grades. Credit assessment is a very important but rather complicated process for commercial banks.From the past decades to nowadays,different schemes using simple financial ratios and many other statistical methods,such as logistical regression analysis,linear regression analysis,decision trees and so on,are used popularly by commercial banks all of the world. With the development of artificial intelligence technology,the method of neural networks has also been applied in this field recently by researchers,and their findings indicate that neural networks are significantly better than statistical methods in prediction and robust to sampling variations.To perform more efficiently,a Mixed Expert System (MES) based method is proposed in our research,which combines Artificial Neural Network (ANN) technique with traditional Expert System (ES) to deal with the qualitative factors of various aspects in enterprise credit assessment. The focus of this work is concentrated at the ANN module of that method.Instead of dealing with the whole sample set all together,we first divide it into several moderate sets,or patterns,based on the industry and possession attributions of the enterprises concerned. When a pattern becomes a reasonable sample sub-set,we develop an ANN model for it. In this work,we have constructed 15 individual ANN structures for all the patterns we have got. The results of our experiment show that it increases the accuracy of the assessments significantly.By adopting object-oriented technology to analysis the global system and its ANN module,and using Borland C++ Builder software to program,an enterprise credit assessment prototype system based on mix-expert system has been developed. We hope that it will be a useful computer-aided tool for the commercial banks.