審計人員及會計師對「繼續經營意見（going-concern opinion）」的判斷是非常關鍵性的，如判斷失誤未能發現企業破產發生之可能性，將造成財務報表使用者以及企業相關的利害關係人（stakeholders）極大的損失，而傳統的統計模式在「繼續經營意見（going-concern opinion）」判斷上有很大的缺點，失誤也較高。迎接大數據時代的來臨，近年來有一些研究使用資料探勘技術來判斷繼續經營疑慮，也降低了判斷的失誤。國際四大會計師事務所(Deloitte, KPMG, PwC, EY)也愈來愈重視大數據與AI人工智慧審計查核。本研究計畫整合數種AI人工智慧與機器學習技術，包括：第一階段以分類迴歸樹(CART)、卡方自動交叉驗證(CHAID)及QUEST三種決策樹之演算法、類神經網路、約略集合等資料探勘技術篩選出重要變數，第二階段配合資料探勘中的支援向量機與貝氏認知網路(bayesian belief network, BBN)以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，建構有效（能更準確預測）的繼續經營審計意見決策模型。本研究之研究對象為2000年至2019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上櫃公司。
“Going-concern opinion” is a very critical judgment for the auditors and accountants, such as fail to find the bankruptcy of enterprises, and will result in a great loss for their stakeholders. There are lots of shortcomings on the traditional judgment model on “going-concern opinion”. The international big four accounting firms (Deloitte, KPMG, PwC, EY) also pay more and more attention to big data and AI artificial intelligence audits. This project try to use artificial intelligence (AI) and machine learning techniques to improve the going-concern prediction accuracy for CPAs. In the first stage, the decision trees of three algorithms, including the Classification and Regression Trees (CART), Chi-squared Automatic Interaction Detector (CHAID), QUEST, Artificial Neural Network (ANN), and Rough Set (RST) will be applied in the selection of major variables. The second stage will combine the Support Vector Machine (SVM), Bayesian Belief Network (BBN), and Artificial Neural Network (ANN) to establish classification models for comparison. Both financial and non-financial variables were used to establish the prediction models. This study project will adapt financial and non-financial variables for the prediction models. The initial dataset comprises companies listed in the Taiwan Economic Journal (TEJ) from 2000 to 2019.
