Intelligent High Dimensional Data Mining Systems for Financial Distress

• Main Authors: Yi-Ching Lee, 李怡靜
• Other Authors: Shian-Chang Huang
• Format: Others
• Language: zh-TW
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/66634596106214415606

碩士 === 國立彰化師範大學 === 企業管理學系 === 100 === In recent decades, bankruptcy prediction is an indispensable issue in the area of academia and industry. Investors can get the information from the financial statement to analysis that the company's capital structure and ability of making profit when it faced with financial crisis. However, most of the early warning models were based on financial ratios and corporate governance as indicators in previous research. In this study, we devises a financial distress prediction model not only takes financial indicators and corporate governance into consideration, but also includes macroeconomic factor, Audit Opinions, Auditor Changes and audit firm changes as the explanatory variable. And we examine a sample of the Taiwan Stock Exchange Corporation (TWSE) from 1999 through 2010.In addition, this paper which examines ratios with 2^111 high dimensions for data mining. However, such data of high dimensions could decrease the performance of accuracy. Therefore, the aim of this study is trying to use the KLFDA method to reduce dimensions in order to feed ensembles. Second, the underlying classifiers we used are logistic regression, support vector machine, neural networks, C4.5, K-NN and bayesnet. The ensemble algorithms used are Adaboost M1, Bagging, Dagging and Random Subspace, and then we compare underlying prediction algorithms from data mining and ensemble methods. At last, comparing with all of the ensembles, and finding out the best model for financial early warning systems. Empirical results indicated that integrating KLFDA with IBK-based Dagging achieves best accuracy