Modeling consumer loan default prediction using neural netware

Abstract

In this paper a loan default prediction model was constricted using two attribute detection functions, resulting in two data-sets with reduced attributes and the original data-set. A supervised two-layer feed-forward network, with sigmoid hidden neurons and output neurons is used to produce the prediction model. Back propagation learning algorithm was used for the network. Furthermore three different training algorithms were used to train the neural networks. The neural networks are trained using real world credit application cases from the German bank datasets which has 1000 cases; each case with 24 numerical attributes; upon which the decision is based. The aim of this paper was to compare between the resulting models produced from using different training algorithms, scaled conjugate gradient backpropagation, Levenberg-Marquardt algorithm and One-step secant backpropagation (SCG, LM and OSS). This study show that although there is no great difference between LM and SCG but still LM gives better results. The attribute reduction function used helped to produced models quickly and more accurately.