Credit scoring using neural networks trained with augmented discretized inputs

Abstract

We apply neural networks that have been trained and pruned using augmented discretized input data for credit scoring. Credit scoring datasets normally contain input data attributes that are continuous (e.g. salary) and discrete (e.g. marital status). In order to improve the accuracy of the neural network prediction, we augment the input data by including the discretized values of the continuous attributes. Having both the original continuous attributes and their discretized values make it easier for the networks to form decision boundaries that could be axis-parallel or oblique in the input space defined by these attributes. Neural network

5 Figures and Tables

Cite this paper

@inproceedings{Azcarraga2013CreditSU, title={Credit scoring using neural networks trained with augmented discretized inputs}, author={Arnulfo P. Azcarraga and Yoichi Hayashi}, year={2013} }