由於台灣經濟衰退和失業率上升，除了對信用卡產業造成很大的負面影響外，也為信用卡發卡機構的違約品保分析師及管理經理帶來很大的壓力，迫使他們必須採取有效措施以防止不斷上升的違約率。本文應用灰色關聯分析建構一兩階段的信用卡違約預測模型，在第一階段的特徵選取中應用證據推理方法來結合不同灰色關聯分析所產生的變數排序，透過資料融合為一最佳排序來挑選重要變數。在第二階段的違約案例預測則採用動態權重的方式來校正並取代原有灰關聯度計算所使用的平均權重。實證結果顯示，在各項模型績效評估指標中以兩階段違約預測模型的準確度最高，其預測準確性可以顯著提升到89.7%。證據推理除了可以減少不同灰色關聯分析所產生的不一致結果，同時也可以提供一種有效的折衷解決方案進行特徵選取。此外利用訓練樣本的機率分佈來計算每一筆測試樣本進行樣式比對時所需的權重，可以有效地動態調整灰色關聯分析提升預測品質。此兩階段違約預測模型能夠協助金融機構違約品保分析師及管理經理改善與違約個案執行相關的品管與稽核流程。

Both the economic downturn and increasing unemployment rate in Taiwan have given rise to a number of negative impacts on the credit card industry and put pressure on the default managers and default quality assurance analyst within a financial institution to take effective measures against the rising default rate. This paper proposed a predictive model of two-stage grey incidence analysis (GIA) which applied evidential reasoning approach to combine various attribute rankings of GIAs in feature selection at the first stage, and then calibrated the instance predictions of GIA with dynamic weighting method for the classification of delinquency cases at the second stage. The experimental results indicated the predictive model performed better in most evaluation metrics and its accuracy could be significantly improved to 89.7%. The evidential reasoning provided an efficient compromise solution in feature selection. Moreover, the GIA model for optimal instance matching was effectively adapted by the dynamic weights calculated for testing samples based on the probability distribution of training samples. The two-stage predictive model will be able to assist default managers and analysts in performing quality control audits to the default related processes.