查核人員若未即時發現企業破產發生之可能性，將造成財務報表使用者以及投資大眾極大的損失。過去繼續經營決策意見的相關文獻中，多以傳統的迴歸模式為主，近年來有許多學者使用資料探勘法進行繼續經營決策意見相關研究，都有相當不錯的準確率，但整體文獻還不夠完整。
故本研究第一階段以資料探勘中的分類迴歸樹(CART)、卡方自動交叉驗證(CHAID)及約略集合(rough set)，三種資料探勘法篩選出重要變數，配合決策樹C5.0、支援向量機與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，希望能得出一個更精準的繼續經營意見決策模式。本研究之研究對象為2004年至2013年間，有被出具繼續經營疑慮意見及未被出具繼續經營疑慮意見之上市及上櫃公司。研究結果發現，分類迴歸樹篩選變數搭配類神經網路之分類效果最好，準確率達95.08%。
If auditors are unable to identify the possibility of bankruptcy occurring to en-terprises in time, financial statement users and investors are likely to ignore the risks and eventually suffer from tremendous loss. According to the literatures reviewed by this study, the great majority of going-concern opinions made in the past were for-mulated using the conventional regression models. In recent years, many researchers studied going-concern opinions using data mining method with satisfactory accuracy. However, not enough literatures are available at this moment.
Therefore, this study attempted to identify critical variables using three data mining methods, namely, Classification and Regression Tree (CART), Chi-square automatic Interaction Detection (CHAID), and Rough Set. This study constructed a classification model and compared data using decision tree C5.0, support vector ma-chine, and artificial neural network. Secondly, this study adopted financial variables and non-financial variables in order to obtain an going-concern opinion model that is more accurate than others. This study focused on the TSC corporations and OTC corporations with or without suspicious opinions regarding their going-concern ex-pressed by auditors in 2004-2013. According to the research results, when variables are selected using CART and artificial neural network, 95.08% of accuracy was ob-tained.
