多分類器組合應用於財務危機預測;Financial distress prediction based on multiple Classifiers

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/57785

Please use this identifier to cite or link to this item: https://ir.lib.ncu.edu.tw/handle/987654321/57785

Title:	多分類器組合應用於財務危機預測;Financial distress prediction based on multiple Classifiers
Authors:	楊順雄;YANG,SHUN-HSIUNG
Contributors:	資訊工程學系
Keywords:	財務危機預測;多分類器;資料探勘;機器學習;Data Mining;Financial Distress Prediction;Machine Learning;Multiple Classifier
Date:	2012-10-04
Issue Date:	2012-11-12 14:36:56 (UTC+8)
Publisher:	國立中央大學
Abstract:	近年來，由於全球金融危機的爆發，許多經濟實體發生都會遭受巨大的損失，甚至導致破產，所以財務危機預測financial distress prediction（FDP）的問題一直以來都是被廣泛討論和持續研究的主題。在企業融資領域，要如何有效地預測財務危機，對於企業財務管理是一個很重要的問題。最近，使用多分類器組合來解決FDP問題越來越被重視。而本研究中就是在多分類器組合的基礎架構上提出了可提升FDP效果的方法。一般典型的財務預測基本上就是做二元分類（危機類和非危機類）的問題。然而，我們發現台灣證券交易所（TWSE）不僅定義了一間公司是否有發生財務危機，且還定義一間公司發生了哪種財務危機。因此，我們想試著建構各類財務危機的專精分類器再利用OR組合機制將所有危機種類分類器的預測結果做組合。我們會在本論文中會提出一個新的多分類器組合架構，並會在論文中試著證明我們所提出的預測模型相較以往傳統的預測模型有著更好的預測表現。In recent years, many economic entities have suffered great loss or even become bankrupt due to the breakout of global financial crisis. Financial distress prediction (FDP) has been widely studding. In the field of corporate finance, to predict financial distress effectively is an important issue in corporate financial management. Recently, some studies which take advantages of multiple classifiers combination to solve FDP problem have been noticed. This paper proposed a FDP method with increased efficiency based on parallel combination of multiple classifiers. Traditional approaches usually take financial prediction problem as Binary Classification (distressed class and non-distressed class) problem. However, we found that Taiwan Stock Exchange Corporation (TWSE) not only defined two classes for all companies but also defined subclasses for the distress companies. Hence, we wanted to construct specific classifier for each distress type and then combine all distress classifiers prediction results with the OR mechanism. We also proposed a new prediction model with multiple classifiers and showed that our method had outperformed the traditional approach.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	653	View/Open

社群 sharing

Loading...