分享:
摘要
近年來隨著舞弊案件不斷增加,當公司發生重大弊案時,不僅使公司本身受到傷害,更造成投資人的重大損失,使得社會必須極大的成本來彌補其所造成的傷害。過去有關企業舞弊的文獻中,主要使用傳統的迴歸模式為主,而近年來有許多學者使用資料探勘來偵測企業舞弊,都獲得相當不錯的準確率,但整體文獻還不夠完整。故本研究第一階段以傳統的逐步迴歸法和資料探勘中的卡方自動交叉驗證(CHAID)及約略集合篩選出重要變數,配合決策樹C5.0及類神經網路分別建構分類模型並進行比較,變數方面則採用財務及非財務變數,希望能建立一套更為有效的企業舞弊偵測之工具。本研究之研究對象為2003年至2013年間,41家發生企業舞弊及123家非企業舞弊之公司。研究結果發現,財務及非財務資訊皆能有效的辨別企業舞弊;且以約略集合篩選變數搭配類神經網路之分類效果最好,準確率達88.46%。故本研究認為約略集合搭配類神經網路之企業舞弊偵測模型能作為協助審計人員於查核過程中偵測企業舞弊之工具,及提供審計人員及投資大眾作為重要決策之參考。
關鍵字:舞弊,資料探勘,決策樹,約略集合,類神經網路,決策樹C5.0 (Decision Tree C5.0),決策樹CHAID (Decision Tree CHAID)
Abstract
More and more corporate frauds have been discovered in recent years. When a corporation is involved in frauds, the corporation and its investors fall victims to all such frauds. Consequently, the society incurs huge costs to make up for the loss caused by corporate frauds. The great majority of literatures adopted regression models to analyze corporate frauds. In recent years, however, researchers began to trace corporate frauds using data mining method with satisfactory accuracy. Unfortunately, not enough literatures are available at this moment. Therefore, this study attempted to identify the critical variables related to corporate frauds using the conventional stepwise regression method, Chi-square Automatic Interaction Detection (CHAID), and rough set. This study constructed an effective tool, using financial variables and non-financial variables to trace frauds. This study focused on 41 enterprises involved in corporate frauds and 123 enterprises not involved in corporate frauds in 2003 – 2013. According to the research results, both financial information and non-financial information were sufficient to reveal corporate frauds. When variables were selected using rough set and artificial neural network, 88.46% of accuracy was obtained. Therefore, this study concluded that the fraud detection model constructed with artificial neural network and rough set is sufficient to trace corporate frauds, and is a perfect tool that helps auditors and investors to reach major decisions in connection with investment or disinvestment.
Key words: fraud, data mining, decision tree, rough set, artificial neural network, C5.0 (Decision Tree C5.0), CHAID (decision tree CHAID)