Biological plausibility in an artificial neural network applied to real predictive tasks

Abstract

Biologically plausible artificial neural networks represent a promising novel approach in bio-inspired computational systems. In these systems, the models are based on existing knowledge of neurophysiological processing principles. Research in this field has increased in the last few years and has generated new viewpoints, propositions and models that are closer to the known features of the human brain. Some researchers have recently focused their studies on this innovative field in order to establish a consensus on what an artificial neural network is in the domain of biological realism. Domain specific synthetic data sets are generally used in the evaluation of those artificial neural networks because they simulate predictive tasks and potential problems caused by human intervention. This paper deals with the analysis of influence of the anomalies generated by human intervention in credit approval process. Such anomalies modify real classification, performance and accuracy. In this analysis, we evaluated a real data set that represents human actions over personal credit approval and fraud identification by using a biologically more plausible artificial neural network proposal. © 2013 Springer-Verlag.