以資料探勘及多重分類器技術建構企業財務危機預警模型

Using Data Mining and Multi-classification Technologies to Construct Corporate Financial Distress Prediction Models

鍾孟杰(Meng-Chieh Chung)

指導教授 ： 李維平

中原大學/商學院/資訊管理研究所/碩士(2014年)

https://doi.org/10.6840/cycu201400590

摘要

近年來，企業經營環境隨著資訊全球化時代的來臨而有了重大的轉變，在整體經濟情勢的愈加困難下，財務危機發生的可能性已隨之逐年的增加。對企業的投資者而言，企業能否繼續經營將是他們是否願意將資金投入資本市場的主因。而企業財務危機更是攸關著企業生存與否最重要的關鍵點，因此若能及早預測出企業財務危機將能減少對企業甚至社會大眾的損失，故企業財務危機預警模式逐漸發展起來。因此，建立一個有效的財務危機預警模型，是當前學術界與實務界的一個重要課題。
　　從過去研究中可以發現，資料探勘所建構的模型優於傳統統計模型，其中又以決策樹與類神經網路模型最受歡迎，此外，近期也有學者將眾多分類模型整合，建構出多分類器預警模型，在成效上也多有改良。但我們認為，此領域仍有再改善的空間；基於以上議題，本研究將提出單一分類器、多重分類器、混合分類器等三大類預警模型，並使用多種分類技術，如：決策樹、類神經網路、最近鄰居法、隨機森林等方法，再結合資料抽樣的Bagging技術，建構多套財務危機預警模型，並綜合分析其預測效果。而在實驗測試環境，我們選用大多學者認定的美國加州大學爾灣分校(University of California at Irvine, UCI)資料庫中的企業資料，期望透過更完整多元的財務危機預警模型，提供企業界及學術界的後續研究依據。

關鍵字

資料探勘 ； Bagging ； 財務危機 ； 多分類器

並列摘要

In recent years, the business environment, with the advent of the information age of globalization while there have been major changes in the overall economic situation more difficult, the likelihood of financial crises have been followed by increases year by year. Corporate investors, companies will be able to continue to operate if they are willing to put money into the capital markets the main reason. The enterprise financial crisis is at stake with the company's most important key point to survive or not, so if the financial crisis early to predict the business will be able to reduce the loss of business and even the general public, so enterprise financial crisis mode gradually developed. Therefore, the establishment of an effective early-warning model of financial crisis, is a current academia and practitioners important issue.
Research from the past can be found, data mining models constructed superior to traditional statistical models, among which the decision tree and neural network models of the most popular, in addition, there are scholars of many recent classification model integration, to construct a multi-classifier warning model, also made many achievements in the improvement. However, we believe that this area is still room for further improvement; Based on the above issues, this study will propose a single classifier, multiple classifiers, hybrid classifier other three categories warning model, and use a variety of classification techniques, such as: decision trees, class neural networks, nearest neighbor method, random forests and other methods, combined with data sampling Bagging technology to construct multiple sets of financial crisis early warning model and comprehensive analysis of the predicted effect. In the experimental test environment, we use most of the scholars identified the University of California, Irvine (University of California at Irvine, UCI) database of corporate information, hope that through a more complete model of diversified financial crisis early warning, providing business and academic community based follow-up study.

並列關鍵字

Bagging ； Multiple classifiers ； Financial Crisis ； Data mining

參考文獻

王美慧（2005）。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，1，1-30。

朱玉婷（2006）。探討電子業及傳統產業於破產預測差異之研究。中原大學資訊管理研究所碩士論文。

李金鳳、蔡孟哲（2001）。利用決策樹分類法建置資料倉儲中檢核與整合大量企業資料之機制，朝陽學報，6，147-166。

李際偉、武彤、郭定謀（2011）。輸入型供應鏈風險分析，運輸計劃季刊，40，185-212。

林郁翎、黃建華（2004）。考慮公司治理之企業財務危機預警模型，東吳經濟商學學報，64，23-56。