A Comparison of Growing Cell Structures Neural Networks and Linear Scoring Models in the Retail Credit Environment

In this paper we address the primary problem of lenders: how to distinguish between low- and high-risk debtors prior to granting credit. We constructed two types of credit scoring models: one based on standard logistic regression and the other on growing cell structures (GCS), something which has never before been applied to this area of research. The credit scoring task was performed on the pooled data set from a new member country of the European Union and its Economic and Monetary Union. The proposed model outperformed the benchmarking (logit) model. The results also demonstrate that the GCS model has better capability of capturing nonlinear relationships among variables and can better handle the properties of categorical variables.