本文旨比較六種資料探勘方法在股票上市公司之財務危機建模的適用性。結果發現，就CAP曲線圖的面積率而言，邏輯斯迴歸、判別分析、類神經網路表現很好；貝氏分類、分類樹表現較差；最近鄰居分類表現最差。但就風險管理的觀點來看，預測模型推估的預測機率是否能代表財務危機機率比分類正確與否更重要。但因為每一筆資料的已知結果不是有，就是無財務危機，其財務危機機率是未知的，因此本文提出一個創新的排序平滑法來估計每一筆資料的財務危機機率。再經由預測財務危機機率(x)與財務危機機率(y)的線性迴歸分析( ŷ= âx + □)顯示，類神經網路產生的預測模型具有最高的判定係數，且只有其迴歸係數â值接近1，□值接近0。因此，類神經網路是六個方法中唯一可以準確估計財務危機機率的方法。

This research aimed at comparisons of six data mining methods for the financial distress of company in the stock market. Based on the area ratio of the CAP curve, the results showed that logistic regression, discriminant analysis and artificial neural networks are more accurate; naïve Bayes classifier and classification tree are less accurate, and k-nearest-neighbor is the least accurate. But based on viewpoint of risk management, whether the forecast probability estimated by the predict model can represent the financial distress probability is more important than the classification error rate. Using the regression analysis (ŷ = âx + ,□) of forecast probability (x) and financial distress probability (y), the predict model produced by artificial neural network has the highest determination coefficient, and the regression coefficient "â" of regression formula based on forecast probability estimated by artificial neural network is close to 1, and "□" to 0. Therefore, in these six data mining methods, artificial neural network is the only method which can accurately estimate the financial distress probability.