Genetic-Neural and Neural-Fuzzy Approaches Applied in the Evaluation of the Mortgage Loan

• Main Authors: Ruey-chuan Shieh, 謝瑞川
• Other Authors: Tsung-Nan Chou
• Format: Others
• Language: zh-TW
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/61301867695404844853

碩士 === 朝陽科技大學 === 財務金融系碩士班 === 92 === The bank credit and loan are the major functions and important source of profit for the commercial banks. However, the stringent competitive environment of banking has forced many commercial banks to transfer their focus from business finance to consumer finance gradually. The mortgage loan management becomes an important issue for commercial banks. Most research in the study of bank credit and loan applied statistical Logist model, Probit model and AHP method. Artificial intelligence approaches such as neural network, fuzzy inference, evolutionary method and gray theory are also involved in this research field recently. The aim of this study is to integrate neural network, fuzzy theory and genetic algorithm together and construct four different models to analyze and evaluate mortgage loan and reduce its risk. The models built in this study including neural-fuzzy, general fuzzy, genetic-neural network and general neural network. The experimental results of this study are concluded as following: 1.The prediction performance of the genetic-neural approach is superior to that of general neural network approach. 2.The prediction performance of the neural-fuzzy approach is superior to that of general fuzzy inference approach. 3.The neural-fuzzy approach acquired the best performance among the four models. 4.The sensitivity analysis of the variables revealed that there were negative influences from annual income and guarantor number. Other variables were positive influences as the previous perdition.