鑑於過往發生過多次全球財政危機讓投資人損失慘重，本研究旨在利用機器學習技術中的類神經網路、區別分析、貝氏網路以及決策樹C5.0等方法，嘗試建立繼續經營預測模型。本研究之樣本選取取自台灣經濟新報資料庫(TEJ)，研究對象為2010年至2019年之上市上櫃及下市下櫃公司，採用以一家繼續經營存有疑慮之公司及與該樣本相同年度及產業別的正常企業分別以1:1及1:3的方式進行配對。實證結果顯示無論以1:1或1:3的比例進行比對，決策樹C5.0的平均準確率皆為三種研究方法中最高，分別為87.98%與86.98%。
Due to investors have suffered heavy losses from multiple global financial crises in the past, the aim of this research is using machine learning such as artificial neural network, bayesian network, decision tree C5.0 and discriminant analysis etc to construct a going concern prediction model. The samples of this study were selected from Taiwan Economic Journal (TEJ), targeting both the listed and unlisted companies between 2010-2019. We compared a company who has concern in continuous operating the business with a normal company in the same year same industry in the ratio of 1:1 and 1:3. The results of research show that regardless of which ratio to apply, the average accuracy of decision tree C5.0 are 87.98% and 86.98% respectively, which is the highest among the three research methods.
