Credit scoring and reject inference with mixture models

Abstract

Reject inference is the process of estimating the risk of defaulting for loan applicants that are rejected under the current acceptance policy. We propose a new reject inference method based on mixture modeling, that allows the meaningful inclusion of the rejects in the estimation process. We describe how such a model can be estimated using the EM-algorithm. An experimental study shows that inclusion of the rejects can lead to a substantial improvement of the resulting classification rule. Copyright © 2000 John Wiley & Sons, Ltd.