建構企業危機預警模式已是學術與實務界中長久討論的議題,過去有關企業危機的研究指出人工智慧如支持向量機的優異能力。而隨機森林為Breiman所發展出的新興分類方法,儘管各領域中都有應用隨機森林的研究,也有良好的分類結果,但在財務領域的研究中,隨機森林的應用並不廣泛。因此,本研究提出一整合流形學習與隨機森林的兩階段模式建構程序,來進行企業危機模式之建立。主要研究的程序,首先將輸入資料經由流形學習方法進行資料降維,再將降維之結果當作隨機森林之輸入資料,以利於隨機森林分類正確率之提升,藉此發展一個更為準確的企業危機預警模式。實證結果顯示本研究所提之方法其分類正確率優於單純使用隨機森林和支持向量機,以提供企業或投資者事前洞悉企業危機的徵兆與投資判斷之依據。
To construct of the business distress detection model has long been regarded as an important and widely studied issue in the academic and business community. Nowadays, there have been many successful applications of support vector machines (SVM) in business distress detection and data mining problems, where SVM-based models frequently received state-of-the-art results. Recently, random forest (RF) developed by Breiman, have gained popularity due to many attractive features and excellent generalization performance on a wide range of problems. Nevertheless, there has been little work on the application of RF for finance literature. In this paper, we propose a novel model to integrate manifold learning with RF technique, to crease the accuracy for the prediction of business distress. By manifold learning techniques, we can reduce this high dimensional business distress data into a much lower dimensional space, is utilized as a preprocessor to improve business distress prediction capability by RF. The results show that the proposed model provides better classification results than pure RF and support vector machine, can help investors in correct investment decisions.