近年來，受到全球經濟衰退的影響，公司爆發財務舞弊情況日漸增加，審計品質之議題也逐漸受到各界關注。因此，在變化快速的經濟環境裡，如何建立一個有效的偵測模型，以提供公司管理階層、投資人或相關利害關係人等，正確的評估審計品質，成為一項重要課題。約略集合理論是一種新穎的人工智慧演算法，其適用於處理模糊及不精確的訊息，在高維度複雜資料的運算中，也有相當優異的表現。基此，本研究以約略集合理論為方法基礎，並結合整體學習法來建立審計品質偵測模型。透過兩種方法的結合，能提升模型之偵測效力，並有效縮短運算時間。本研究以2011年到2013年之台灣上市上櫃電子業為研究對象，實證結果顯示，相較於傳統羅吉斯迴歸及單一機制約略集合所建立的偵測模型，混合機制的偵測效果表現較為優異，也證實本研究之研究結果能改善傳統或單一機制之不足之處，並做為評估審計品質之參考。
Auditing quality assessment is an essential topic in recent years when the nu-merous financial and economic scandals burst out and global economic goes into de-pression. Thus, the aim of this study is to introduce an emerging mechanism for re-lated parties such as bankers, investors, stakeholders, to detect the auditing quality so as to make a reliable judgment in highly turbulent economic climate. The detecting mechanism consists of feature selection, detecting model construction and ensemble learning theory. The feature selection is implemented to overcome the curse of di-mensionality. The original data undergoes feature selection procedures will increase the detecting performance as well as decrease the computational burden. Rough set theory with superior ability to handle the vagueness and imprecision was conducted to construct the detecting model. The fundamental idea of rough set theory with en-semble learning theory is used to complement the error made by singular mechanism. The real-life samples ranged from 2011 to 2013 were gathered to examine the effec-tiveness of introduced mechanism. According to empirical examination, the intro-duced mechanism is a promising alternative to assess the auditing quality.
