A number of credit-scoring models that accurately classify consumer loan applications have been developed to aid traditional judgmental methods. This study compares the performance of multiple discriminant analysis (MDA) and neural networks in identifying potential loan. The neural network models consistently perform better than the MDA models in identifying potential problem loans. To alleviate the problem of bias in the training set and to examine the robustness of neural network classifiers in identifying problem loans, we cross-validate our results through seven different samples of the data.
