- 學位論文
利用資料探勘技術建構整合型信用評等最佳化模型
Using Data Mining Approaches to Construct an Integrated Credit Scoring Optimization Model
摘要
信用卡產業面臨強烈競爭與呆帳問題，故信用評等模型的重要性日益增加。近幾年，許多相關領域，例如統計(Statistics)、機器學習(Machine Learning)、人工智慧(Artificial Intelligence)，紛紛投入信用評等的研究。本研究目的為減少人為因素、節省作業成本及加強信用卡授信風險管理，故建立一套客觀、快速的信用卡授信方法與開發潛在客戶，以增加銀行收入並降低潛在之信用風險。然而，不同分類器在各種情況下會有不同的表現，沒有任何一種分類器其分類結果永遠最佳，故本研究提出整合型信用評等最佳化 (Integrated Credit Scoring Optimization, ICSO) 模型，藉以探討何種模式為最佳。本研究將分別進行未篩選變數及利用多元適應性雲形迴歸 (MARS) 篩選變數之模式驗證，其模式包含：(1).單一模式：分為單一信用評等模式與再確認信用評等模式； (2). 多重組合模式：分為不一致信用評等模式與投票機制信用評等模式。(3). 最佳化信用評等模式：最後，利用基因演算法 (Genetic Algorithm, GA) 來找出上述兩種模式之最佳的分類組合，進而從中做出最終之結論。研究結果顯示，ICSO模型可有效提升分類準確率，並同時降低型一與型二誤差，且經過篩選變數後，雖不能明顯提升分類準確率，但仍可有效地減少各分類器在分類時所需要的執行時間。
並列摘要
The use of credit scoring model has become a very important issue as the credit card industry is confronted with ever-mounting competition that in turn triggers serious problems in bad debts. Credit scoring model has therefore been widely studied in the areas of statistics, machine learning, and artificial intelligence in recent years. The purpose of this study is accordingly to construct an objective and fast credit scoring model to help banks minimize the impacts of human factors, develop potential customers, trim operating costs, and reinforce credit risk management. Since different classifiers perform differently in varied situations, the study proposes an Integrated Credit Scoring Optimization (ICSO) Model to examine the performance of various classifiers and explore the best model. This study endeavors to verify the performance of credit scoring based on the original attributes and the one based on the attributes identified by Multivariate Adaptive Regression Splines (MARS). Three models in each approach are examined: (1), The single model that can be further divided into the single credit scoring model and credit scoring reconfirmation model; (2), The multiple combination mode that is further divided into inconsistency-based credit scoring model and voting schemes credit scoring model; and (3), The optimal credit scoring model identified by utilizing Genetic Algorithm (GA) to the previous two models. The research results indicate that the proposed ICSO model is effective in simultaneously upgrading classification accuracy and reducing Type I and Type II errors. Though selection of important attributes leads to no significant increase in classification accuracy, it can be applied to different classifiers to effectively speed up the classification.