Design a Rule Extraction Mechanism of Neural Network for Financial Distress Prediction

• Main Authors: Yeou-wei Cheng, 鄭又瑋
• Other Authors: Haung Jin Sheng
• Format: Others
• Language: zh-TW
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/59153600496422852956

碩士 === 國立雲林科技大學 === 財務金融系碩士班 === 95 === Although neural networks have been applied to solve a variety of problems such as classification and prediction, they are often regarded as black boxes. The purpose of this paper is set up a model of financial distress prediction in accordance with corporate governance. The definition of the financial distress company of research is the listed company which is full-cash delivery stock according to Taiwan Stock Exchange Corporation between November 2004 and January 2007.The matching samples were chosen on one-to-one basis proposed by Beaver(1966)、Altman(1968).First, corporate governance variables are divided into four aspect to select, and these variables were training and pruning via neural network. Second, we extract rule that can explain how the problem are solved proposed by Rudy Setiono et al.(2002). Hence investors and decision maker can use these rules to judge a company is in financial distress situation or not. The result of this paper shows that rule extraction has good prediction rate on the financial crisis problem, wholly classify correct rate is up to 87.5%. Compared with the prediction result of data mining from original data, with the prediction result of rule extraction is better. This paper analyses how to help investors set up a good way judging the financial situation of companies.