Further sensitivity analysis of simple evolving connectionist systems applied to the Lincoln aphid data set

Abstract

This report presents two further experiments over the Aphid data set. The first is a more detailed investigation of the sensitivity of Simple Evolving Connectionist System (SECoS) networks to the exclusion of various combinations of inputs. This is in contrast to the previous work, where only the effect of excluding single variables was investigated. The second experiment investigates a hypothesis that attempts to explain the results found in the first experiment.