somayeh mousavi - Ayatollah Haeri University of Meybod, Yazd, Iran

Credit risk measurement and management is one of the most important and critical subjects in bank lending decisions. Credit scoring models are used to distinguish good loan applicants from bad ones. So far, a lot of classical and artificial intelligence techniques are applied in credit scoring applications. According to the previous researches, the artificial intelligence techniques outperform the others significantly. From another point of view, credit scoring models can be categorized into two groups; function-based and induction-based methods. Although the fimction-based methods are powerful in extracting accurate scoring models, however they suffer from interpretability issues. In this paper genetic programming, as an artificial intelligence technique, has been extended to integrate the advantages of function-based and induction-based approaches in one credit scoring model. Thereafter, an artificial neural network-genetic algorithm hybrid system is proposed as an alternative for the previous model. The proposed models are implemented on a well-known real world credit application cases from the German credit data set. Comparison results show the proposed genetic programming model is superior to the other models have been applied on the same data set.

credit scoring; genetic programming; artificial neural networks; genetic algorithm; function-based model; induction-based model