混合模型應用於貸款違約預測

Abstract

在金融市場與貸款活動的發展下，對金融機構和借款人來說，預測貸款違約的機率是個極為重要的研究議題。本論文使用群聚演算法將資料根據相似性分成不同的群組，再依照分群結果建立分類器。根據實驗結果顯示，基於此種方式建立的預測系統可以有效的提高預測違約風險的準確性。另外本論文會透過選擇不同的特徵來分析其對違約預測的影響，發現同時使用借款人資料和貸款相關資料可以取得較好的表現。

With the development of financial markets and lending activities, predicting the probability of loan default has become a crucial research topic for financial institutions and lenders. This thesis utilizes clustering techniques for pre-classification and subsequently develops a classifier. The resulting system enhances the accuracy of default prediction. Additionally, this thesis examines the impact of different features for the prediction of default and finds that simultaneously utilizing borrower information and loan-related data achieves better performance.