運用資料探勘技術建構企業財務危機預警模型

Applying Data Mining Techniques to Construct Forecasting Model of Enterprise Financial Crisis

蔡永泰(Yung-Tai Tsai)

指導教授 ： 李維平

中原大學/商學院/資訊管理研究所/碩士(2005年)

https://doi.org/10.6840/cycu200500626

摘要

企業財務危機預警模式的研究，長久以來，一直是政府機關、金融業者、企業單位及投資者所關注的課題，因此有為數眾多的學者在這個問題上，提出了不同的企業財務危機預警系統，來降低利害關係人的損失。回顧過去預警模式的相關研究，諸多的文獻證實決策樹與類神經網路模型的非線性模式顯然優於傳統統計模型的線性模式，因此決策樹與類神經網路模型已普遍被學者採用。然而，類神經網路在學習上的收斂表現，常會受到是否有適當的輸入變數所影響，所以過去學者曾結合不同的輸入變數篩選方法，來提高類神經網路在預測上的準確率，而近來被用來建構財務模型的決策樹也被用來當作篩選的方法。然而過去學者的研究結果顯示，決策樹作為類神經網路在輸入變數上的篩選，並沒有提高類神經網路在預測上的準確率。本研究深入分析決策樹所篩選之輸入變數，為何無法提高類神經網路預測準確率的原因，並從中衍生出有用的輸入變數篩選方法，並結合類神經網路建構本研究之企業財務危機預警模式。研究藉由比較其他不同的預警模式，以及不同的輸入變數篩選方法，以驗證本研究的可行性。
實驗資料顯示，結合本研究的輸入變數篩選方法與類神經網路所建構出的預警模式，在準確率的表現上優於決策樹模式、類神經網路模式與結合決策樹為輸入變數篩選之類神經網路模式。證實本研究所提出輸入變數篩選方法，能提高類神經網路預測的準確率。另外，實驗資料也顯示，本研究的輸入變數篩選方法優於因素分析法、逐步迴歸分析法與敏感度分析法。

關鍵字

敏感度分析法 ； 類神經網路 ； 因素分析法 ； 逐步迴歸分析法 ； 決策樹

並列摘要

The research of the early-warning mode of the enterprise finance crisis, for long time, have been the topic that the government agency, financial operator, the enterprise unit and investor pay attention to, so have for on this problem, numerous scholars of number put forward the early-warning system of the different enterprise finance crisis, reduce the relation person's loss. The review passes by the related research of the early-warning mode, the many documents confirms the decision tree and a neural network model is not the line mode surpasses the line mode of the tradition statistics model obviously, therefore the decision tree and a neural network model are already widespread scholar adoption. However, a neural network is on the learning of refrain from rash action the performance, often will be subjected to whether have the importation variable of the adequacy the influence, so in the past the scholar has ever combined the different importation variable sieving method, raised a type of neural network on the forecast the accurate rate, and was used recently be also used to come to in the light of method of sieving to the decision tree of constructing the finance model. However the scholar's research result showed in the past, the decision tree was a type of neural network sieving in the importation variable, did not raise a type of neural network in accurate rate on the forecast. The importation variable that the thorough analytical decision tree of this research sieves, why can't raise the reason that a type of neural network predicts the accurate rate, and spread out the useful importation variable sieving method from it, and combine a type of neural network construction studies originally of the enterprise finance crisis of early-warning mode. Study by early-warning mode of other dissimilarities, and the different importation variable sieving method, with identify the feasibility of this research.

Experiment the data manifestation, the importation variable that combines this research sieves the early-warning mode that the method and a neural network construct, surpassing the decision tree mode, a neural network mode and a neural network mode that combine the decision tree as the importation variable sieving on the performance of the accurate rate. Confirming this research puts forward importation variable sieving method, can raise accurate rate of a type of neural network forecast. Besides, experiment the data to also show, the importation variable sieving method of this research surpasses the factor analysis, the stepwise regression analysis and the sensitive analysis.

並列關鍵字

Sensitive Analysis ； Neural Networks ； Stepwise Regression Analysis ； Decision Tree ； Factor Analysis

參考文獻

1.王美慧，「運用類神經網路於證券業網路下單服務品質之研究」，顧客滿意學刊，第一卷第一期，頁1~頁30，2005。

12.黃振豐、呂紹強，「企業財務危機預警模式之研究－以財務及非財務因素建構」，當代會計期刊，第一卷第一期，頁19~ 頁40，2000。

9.陳生祥，「以總體指標與企業財務資料預測企業財務危機」，私立中原大學資訊管理研究所碩士論文，2005。

1.Altman, E. (1968), “Financial Ratios, Discriminant Analysis and the Prediction of Corporate Bankruptcy”, Journal of Finance, September, pp.589－609.

2.B. Back, T. Laitinen, and K. Sere (1996), “Neural networks and genetic algorithms for bankruptcy predictions”, Expert Syst. Appl., pp. 407–413.