A visual method for determining variable importance in an artificial neural network model: An empirical benchmark study.

Although artificial neural networks (ANNs) are an established marketing support tool, they are criticised for their inability to explain their results. The purpose of this study is to demonstrate a visual approach to ANN variable interpretation. Scanner data are used to build a well-known choice model, first as a multinomial logit and secondly as an ANN. Response elasticity graphs are then built for each ANN model variable. These graphs are used to interpret the model variables and are benchmarked against the t-statistics of the logit. The results suggest that modellers using this visual ANN approach can obtain a result that aids in variable interpretation and possibly provides a richer understanding of model behaviour than a standard statistical methodology. [ABSTRACT FROM AUTHOR]