機車貸款業務，貸款人與擔保品都是流動不固定，對於客戶之掌握更為困難，因此貸放前之徵信審查工作與核貸過程是否嚴謹，將更形重要。個案公司長期經營機車貸款業務，以個案公司之現有客戶為資料來源，希望從個案公司過去客戶群中，曾經發生逾期繳款導致損失之個案，將其作較科學及有系統的分析，找出相關影響因素，作為徵授信之準則與依據以減少逾期案件之發生，因而減少損失同時增加獲利與盈餘。 本研究藉由資料探勘技術，建立一套系統的風險評估流程及授信審查機制，以達到作業標準化、過程效率化之目的，日後不管對核貸的風險管理或授信人員的教育訓練，都有依據的方向，並降低承辦到違約案件的機率。 本研究運用SPSS等公司所提出的資料探勘程序模型(CRIPS-DM)結合資料探勘技術中的分類方法，找出授信戶發生逾期還款的主要特徵及因素並建置預測模型。並以SQL Server 2008資料探勘工具針對個案公司提供之實際資料進行資料分析及建置模型。 結果本研究發現，將四種演算法整合應用投票之預測性正確性及成本評估比較都比單一演算法為佳。Motorcycle loans, lenders and collateral are total current assets to cause the difficulty in grasp of customers, so it will be even more important by rigorous credit review and loan approval process before loan. Case study the company run motorcycle loan business in a long-term period, hope by a scientific and systematic analysis on the case of existing customers as sources of information, to identify related factors from the case base in the past clients have caused the loss of late payment, as the criteria and the basis for levy credit to reduce the occurrence of overdue cases, thus reducing the losses and increase profits and earnings. This research data mining techniques can establish a system of risk evaluation and credit examined mechanism to achieve the purposes of standardized operation and efficiency of process. On loans of credit risk management education and training staff have the basis of the direction and contractors to reduce the probability of default cases. In this study, companies such as the use of SPSS data mining program proposed model (CRIPS-DM) combined with data mining techniques in the classification of credit account to find out the main features of occurrence of late repayment and build prediction models and factors. And to SQL Server 2008 data mining tools provided for the case of real data to analyze the data and build models. The results of this study found that the integrated application of the four algorithms to predict voting comparative evaluation of accuracy and cost is better than a single algorithm.
