Title: 應用增生少數合成技術與二階段邏輯斯迴歸方法建構汽車租賃信用風險評估模型
Construction of Credit Risk Assessment Model for Car Leasing by SMOTE and Two-Stage Logistic Regression Techniques
Authors: 陳威廷
Chen, Wei-Ting
Chang, Yung-Chia
Keywords: 信用風險評估模型;邏輯斯迴歸;增生少數合成技術;類別不對稱;credit risk assessment model;logistic regression;synthetic minority over-sampling technique;category asymmetry
Issue Date: 2017
Abstract: 在汽車租賃業中辦理汽車租賃時,大多經由租賃專員由申請戶所填寫之資料進行與其往來與否的評估標準,以人工判斷的方式不僅費時,也會受到租賃專員個人的專業經驗而影響決策的品質。目前中、外文獻中針對汽車租賃業以量化的方式針對申請戶的信用風險評估相關研究非常稀少,除了資料不易取得外,對可用於衡量汽車租賃業信用風險的變數也少見相關的討論。本研究針對汽車租賃業的信用風險評估模型進行研究,利用增生少數合成技術處理高風險與低風險資料數量比例懸殊的問題,再以二階段的方式,應用邏輯斯迴歸以增強風險評估模型的分類效果。本研究利用以台灣某金融機構所提供之汽車租賃申請案件的實際資料來驗證所提出方法的可行性與有效性。研究結果顯示本研究所提出之方法能夠有效處理資料極端不對稱的問題,並可有效用於客觀地評估汽車租賃申請案件的違約風險。
In traditional car leasing industry, most of the case approvals are conducted by car-leasing appraisers based on the information provided by the applicants. This process is not only time-consuming but also subjective to appraisers’ personal and professional experiences. As a result, it may lead to wrongfully approving a case which turned into default or rejecting a good-quality applicant. Objectively using quantitative methods to measure credit risks of loan applicants by financial institutes has been widely applied and accepted. However, literatures regarding using such methods in making car-leasing decisions are rare which may due to the fact that that real data is not easy to collect and also lack of study in discussing the variables to effectively evaluating the credit risk of car-leasing applicants. As a result, this research aims at assessing the credit risk of car-leasing industry by using quantitative model. This study first applied the synthetic minority over-sampling technique (SMOTE) to resolve the class-imbalance problem found in the data since it was found that the number of good-credit applicants are a lot more than the bad-credit ones. Furthermore, this study designed a two-stage method applied Logistic regression in each stage to enhance the effect of the risk assessment model. A set of real data of car-leasing applications provided by a financial organization in Taiwan is used to demonstrate the effectiveness of efficiency of the propose model. The results shown that SMOTE was more effective than over- or under sampling methods in terms of resolving the class imbalance problem found in the data. Moreover, the variables chosen in this study for model building along with the proposed approach were be able to objectively assess the credit risk of the car-leasing applicants.
URI: http://etd.lib.nctu.edu.tw/cdrfb3/record/nctu/#GT070453305
Appears in Collections:Thesis