Application of support vector machines for reject inference in credit scoring

In the banking industry, credit scoring models are commonly built to help gauge the level of risk associated with approving an applicant. Credit scoring models are built on a sample of accepted applicants whose repayment and behavior information is observable once the loan has been issued. For application credit scoring, any declined applicant did not use anymore in the model, since the observation contains no outcome. However, when an applicant is rejected, there is a probability that he has good behavior, but he is rejected because of a miss classification. That is why the rejected applicant should be reconsidered. It will be useful for increasing the company's market share. Reject inference is a technique used in the credit industry that attempts to infer the good or bad loan status of the rejected applicants. The objective of this research is we want to classify the rejected applicants into 'good' and 'bad' behavior by using Support Vector Machines (SVM). The results are very encouraging, we found that SVM achieved 85% accuracy rate with RBF kernel, 40% data training, and σ = 0.0001.