A Triple Artificial Neural Network Model Based on Case Based Reasoning for Credit Risk Assessment

For most credit risk assessment models, decision attributes and history data are of great importance in terms of accuracy of prediction. Decision attributes can be classified into two types: numerical and categorical. As these two types have different characteristics, there will be interference if they are used simultaneously in the same model. By applying the case based reasoning (CBR) and artificial neural network (ANN), this study attempts to use numerical and categorical attributes separately in different phases application of the model. For example, if numerical attributes are used in CBR to select similar cases, categorical attributes will be used as inputs of an ANN based on the cases selected. Therefore, interference caused by the different types of attributes is avoided and the accuracy is improved. As only similar history data are selected and input in the ANN, accuracy is improved further. With the idea above, a triple ANN-CBR model is designed in this paper. This model synthesizes advantages of CBR and ANN. Practical examples show that the model established in this paper is feasible and effective. Compared with other models, it has a better precision performance.