過去對財務危機預測都是用比較傳統的迴歸模式，近年來有許多學者使用資料探勘方法針對財務危機預測做研究，準確度有提高，但整體文獻還不夠完整，所以本研究以應用資料探勘於財務危機預測，希望能得出一個更精準偵測模式。本研究應用2004年到2014年發生財務危機公司當作樣本以一家財務危機公司對二家財務正常公司的方式進行配對，嘗試以類神經網路、逐步回歸與決策樹C5.0結合較新之人工智慧演算法「決策樹CHAID」、「決策樹CART」來建立一個二階段的財務危機預警模型，並探討不同變數篩選方法與決策樹 CHAID和決策樹CART所建立之模型效率差異。本研究發現，以類神經網路、逐步回歸與決策樹C5.0進行變數篩選結合決策樹 CART將有助於提升模型整體準確率，準確率都達90%以上，三種方法型一的錯誤率為1%，本研究結果顯示以類神經網路、逐步回歸與決策樹C5.0篩選搭配決策樹 CART可以有效地預測企業財務危機。
To predict the financial crisis of enterprise in the past using traditional regression models, many scholars have been using data mining method of forecasting enterprises financial crisis. The accuracy has improved, but it is incomplete in the literature. This study uses data mining to forecast a business's financial crisis. The data is from TEJ during year 2004 to 2014. In the first stage, we using neural network, stepwise regres-sion and decision tree C5.0. In the second stage we combine with two artificial intelli-gence algorithms “ decision tree CHAID ”, and “ decision tree CART ” to establish two stages of financial crisis predicting models. The results show that NN+CART, SR+CART and C5.0+CART have the highest accuracy of 90%.
