企業對於財務困境的評估需求越來越高，因為全球企業的財務困境的案例越來越多，因此對於更有效的財務困境預測模型有更大的需求。機器學習技術在預測問題的應用，主要著重在不同的技術中尋找出一個準確率最高的模型來做為預測。僅使用監督式學習的單一模型，在預測的準確率上已經不容易有突破性的發展，導致出現一個新的趨勢，就是整合多種演算法來增加資料探勘的表現。混合式資料探勘乃是使用兩種以上的學習法的優點藉以提昇單一學習法的效能或效率。隨著混合分類器的漸漸成為研究的趨勢，使用混合式架構，整合多種技術的效果確實比單一分類技術還要來得好。過去的文獻鮮少相關研究探討不同類型的混合模式搭配的預測表現效果如何。 鑑於做出更準確預測的重要性，本研究將要比較機器學習與統計型的分類器搭，配機器學習或是統計型的分群法演算法，比較何種混合模式能提供在財務預測資料集有最準確的預測結果。本研究以四個組合的混合模式在多個破產預測資料集進行實驗，組合為Affinity propagation搭配Support vector machine；K-means搭配Logistics regression；Affinity propagation搭配Logistics regression；K-means搭配Support vector machine。經過實驗以Affinity propagation搭配Support vector machine在較多資料集有最好的預測表現，平均AUC亦是最好；Affinity propagation能幫助Logistics regression提升預測的準確率在要求較低建模時間的情況成為一種選擇。期望研究結果能幫助實際進行建立實務預測模型的參考。 ;This study investigates the efficacy of applying variant hybrid machine learning models to bankruptcy prediction problem. Although it is a well-known fact that the hybrid models perform well in prediction tasks, the method has some limitations in that it is an art to find an appropriate hybrid model structure. Fewer studies explore how the predictive performance of different types of mixed hybrid model collocations. This study will compare machine learning and statistical classifiers with machine learning or statistical clustering algorithms. Four combinations of mixed models were used to perform experiments on multiple bankruptcy prediction datasets. The combination was Affinity propagation with Support vector machine; K-means with Logis-tics regression; Affinity propagation with Logistics regression; K-means with Support vector machine. The results demonstrate that the accuracy performance of Affinity propagation with Support vector machine has the best predictive performance in many datasets.
