 |
An International Meeting of American Accounting
Association
|
An Application of Machine Learning Fuzzy Expert System in Bankruptcy Prediction |
Cher Hsueh - Ju Chen
National Chung Hsing University
Lin Chin - Shien
National Chung Hsing University
Qiu Qiong - Ye
The Providence University
Abstract: Among the methods used in bankruptcy prediction, the traditional statistical method can only explain the linear relationship among the independent and dependant variables, but it cannot explain the possible nonlinear one. The expert system can provide the more detailed rules among the variables; however, it is hard to extract the knowledge base. Although neural network can obtain the complicated mapping function between the variables, it cannot explain the causal relationship among them. Therefore, this paper is trying to construct a bankruptcy warning system by using neural fuzzy, a hybrid approach combining the functionality of fuzzy logic and the learning ability of neural network. The empirical results show that neural fuzzy is superior to logit in terms of its better accuracy rate, lower misclassification cost and higher detecting power. Besides, the obtained knowledge can also provide a more detailed relationship among the variables, which can be further tested.