Neural Network (NN) is one of the most commonly used method in credit scoring. On the other hand Fuzzy Logic (FL) cannot be used alone in credit scoring due to lack of learning ability. However this is possible in hybrid models where NN and FL are used together. This study investigates the possible nonlinear relationship between the characteristics of loan applicant and default probability by using hybrid Adaptive Neuro Fuzzy Inference System (ANFIS) which combines NN and FL. We used statistical techniques such as discriminant analysis, logistic regression and multivariate adaptive regression splines as feature selection before developing credit scoring models. Then we developed 83 credit scoring models by using NN and 15 credit scoring models by using ANFIS. While the optimal ANFIS model has 78.5% classification accuracy, NN has 77.5%. In addition to higher classification accuracy the ANFIS model is not a black box, in contrast to NN. The ANFIS model can explain the reasons for the credit decision. So this study reveals the nature of personal loan decisions with the German credit data set.
- Pricing brazilian exchange rate options using an adaptive network-based Fuzzy inference system
- Behavioristic analysis and comparative evaluation of intelligent methodologies for short-term stock price forecasting
- RBF centers initialization using fuzzy clustering technique for function approximation problems
- Credit card application assessment using a neuro-fuzzy classification system