A Combined Learning Approach for Credit Scoring Using Adaptive Hierarchical Mixture of Experts: Iranian Banking Industry

Traditional methods for granting credit to loan applicants are based on personal judgment. Nevertheless, the current financial crisis alongside the efforts of banks and financial institutes for decreasing the percentage of overdue loans emphasis the importance of Credit Scoring (CS) models. This paper provides a credit scoring model by means of Modular Neural Network (MNN) established upon combined hybrid-ensemble learning. The proposed model is composed of four powerful neural networks that construct collectively the Adaptive Hierarchical Mixture of Experts (AHME). Training process is a hybrid way for learning the modular model and adaption to the CS model based on the modulation of learning rules specific to each module and particular HME online learning algorithm. Binary Particle Swarm Optimization (BPSO), using Taguchi reasoning scheme for tuning the governing parameters, is also applied for reducing dimensionality and decomposing the problem among the various modules. The proposed model’s performance is compared with that of Multi-Layer Perceptron (MLP) and Laterally Connected Neural Network (LCNN) models. The aforementioned models are evaluated using the data obtained from one of the Iranian banks. Results demonstrate that the AHME outperforms other methods in terms of prediction accuracy as well as the Area Under the ROC Curve (AUC) and the Mean Squared Error (MSE) rate.