世界各國政府皆以開始防範盈餘管理行為，但還是會有少數企業逐漸面臨倒閉時，開始進行盈餘管理，管理階層會因此操縱或是改變原先模式而去評估及解決，以減少盈餘管理較嚴重的程度發生以及被外部投資人發現的比率，超過一定程度時便會成為舞弊公司，無論企業內部或是外部皆會走向沒落。本研究選用資料來自於臺灣經濟新報資料庫(TEJ)，選用期間為2010-2017年臺灣電子產業，以往學者多利用迴歸模式作為分析，近幾年學者利用資料探勘模式(data mining)的準確率相對於以往較為提高。故本研究以支援向量機(support vector machine)作為第一階段的變數篩選，選擇對盈餘管理較有重要性之變數，再利用決策樹CART、決策樹CHAID、決策樹C5.0以及決策樹QUEST作為第二階段模型，以相互搭配的方式選擇較為準確的盈餘管理預測模型。實證結果顯示，由支援向量機搭配決策樹C5.0(SVM-C5.0)的模型為最佳預測之模型，準確率為88.11%。
The selection data of this study comes from the Taiwan Economic News (TEJ). The selection period is from Taiwan to the electronic industry in 2009-2017. In the past, scholars used the regression model as an analysis. In recent years, the accuracy of data mining by scholars is relative to that of data mining. In the past, it has improved. Therefore, this study uses the support vector machine as the first-stage variable screening, selecting the variables that are more important for earnings management, and then using the decision tree CART, decision tree CHAID. As the second stage model, decision tree C5.0 and decision tree QUEST are used to select a more accurate earnings management prediction model. The empirical results show that the model of support vector machine matching decision tree C5.0 (SVM-C5.0) is the best prediction model with an accuracy rate of 88.11%.
