企業所發布之財務報表上的攸關性及可靠性與管理階層操縱裁決性應計項目的盈餘管理行為多年一直是重大議題。過往文獻於裁決性應計項目盈餘管理上皆是使用傳統的回歸模式，近年來已有許多學者應用資料探勘的方法針對應計項目盈餘管理進行研究，然而準確度皆有所提高，因此本研究以資料探勘法中的類神經網路與決策樹來進行預測，希望能找出一個較準確地預測模式與規則。本研究嘗試以類神經網路(artificial neural network，ANN)、決策樹CHAID (Chi-Square Automatic Interaction Detector for Decision trees)以及決策樹QUEST (Evaluation in a questionnaire for Decision trees)先將變數進行第一階段的篩選，再進一步使用支援向量機及決策樹C5.0來建立模型檢測企業是否具有嚴重操縱盈餘的情況。而實證結果顯示，決策樹CHAID搭配決策樹C5.0表現最佳，準確率為98.12%。
Companies release and reliability financial statements and management manipulated discretional accruals earnings management of arbitral accruals has been a major issue for many years. Past literature on the discretional accruals earnings management are all using traditional regression model, in recent years there have been many scholars have applied data mining methods for discretional earnings management research. Therefore, this study data mining methods neural network and Chi-Square Automatic Interaction Detector for Decision trees to predict, hoping to identify a more accurate prediction models and rules. This study attempts to artificial neural network (ANN)、Chi-Square Automatic Interaction Detector for Decision trees and Evaluation in a questionnaire for Decision trees first stage of the first variable filter, further use of support vector machine and decision tree C5.0 to model detection whether a company has serious manipulate earnings. The decision tree C5.0 demonstrates the best performance for earnings management with the accuracy rate of 98.12.
