財務危機預測長久以來都是一個重要且常被廣泛討論的主題，發展出好的財務危機預警模型可以有效幫助，銀行決策。影響整個財務危機預警流程主要有兩個議題分別是特徵挑選(Feature selection)與分類器演算法(Classifier algorithm)，過去研究顯示，單純改良分類器演算法，準確率很難有顯著的提升。本論文將目標放在另一個議題特徵挑選，我們觀察到財務比率數量會隨著年代大幅成長，如何在大量的財務比率下挑選出重要的財務比率，變成很重要的議題，近幾年，研究顯示基因演算法應用於特徵挑選在單一特定的資料集下表現相當好，我們知道特徵集合成長速度相當的快速，如果只驗證基因演算法在單一特定的特徵集合的效果是不足夠的，本論文模擬了特徵集合越來越大的情況，觀察基因演算法表現情形，最後觀察出基因演算法在不加入公司治理的特徵情況下，當特徵集合越來越大，基因演算法挑選出來的特徵組合，準確率還是能夠穩定成長而且較其他特徵挑選方法穩定。Financial distress problem has been important and widely studied topic, development of good financial analysis model can help bank to decisions. There are two major factors, namely feature selection and classifier algorithm, influencing financial distressed prediction. Previous researches show that the forecasting accuracy is very difficult to have significant improvement by improving classification algorithm only; therefore, our research focus on the feature selection issue. Over time，we observed financial ratio growing quickly, that mean feature selection become more important, In recent years, Previous researches have shown genetic algorithm applied to feature selection in unique feature set have good performance, but we know feature size growing quickly, it is not enough to prove genetic algorithm in unique feature set. In our research, we simulate ratio growing situation, consider genetic algorithm performance. Finally, if we exclude corporate governance, we discover genetic algorithm predict performance become well when feature size larger.
