Restructuring the credit process: behaviour scoring for german corporates

Abstract

An instrument for automated monthly credit standing analysis based on data of the corporates current accounts is presented. Different methods of statistics and machine learning are used to develop scoring models for the supervision of debtors. The following methods were selected for model developement: Linear Discriminant Analysis Pattern Recognition (k-nearest-neighbours) Genetic Algorithms Neural Networks Decision Trees The developed models were compared not only concerning their classification results but also concerning score distribution, transparency and IT-realisation. Copyright © 2000 John Wiley & Sons, Ltd.