財務危機預測的問題長久以來都是一個重要且常被廣泛討論的主題,吸引了世界各地的投資者和研究學者的關注。發展出好的財務危機預警模型可以有效幫助,投資者銀行決策。影響整個財務危機預警流程主要有三個議題分別是特徵挑選(Feature selection)、分類器演算法(Classifier algorithm)和資料集(Dataset)。過去由前人的研究可發現,是否要做特徵挑選方法會依據分類器的特性來判定。本研究專注在以準確率和Type I作為指標,探討常用的分類器是否需要做特徵挑選,最後推薦一套方法針對未來資料集可以縮小搜尋的範圍和時間。 Financial distress problem has been important and widely studied topic. Financial distress prediction is receiving increasing attention of stakeholders and researchers in the worldwide. It is helpful for stakeholders and researchers that developing a great ?nancial distress prediction model. There are three important factors influencing financial distressed prediction. The first one is feature selection, the second one is classifier algorithm, and the third one is dataset. It can find that it is useful to use feature selection in different kinds of datasets in previous studies. The aim of this research is make sure that what kinds of classifier need to use feature selection to have the better accuracy, and also recommend a way that it can reduce a lots of time to search the better combination of feature selection approach and classifier in the new datasets.