Null-model validation of MLP input contribution analysis in ecology

Abstract

A method is presented for applying a null-model analysis to the verification of the significance of the input neurons of Multi-Layer Perceptrons (MLP). This method was applied to a problem from ecology, namely the establishment of invasive insect pest species. Previous work has described how the MLP were trained to predict species establishment from climate data, and to identify which climatic factors are significant. The null-model analysis method described here was used to validate these predictions.