A visual analysis of learning rule effects and variable importance for neural networks in data mining operations

Abstract

This study demonstrates two visual methodologies to support analysts using artificial neural networks (ANNs) in data mining operations. The first part of the paper illustrates the differences and similarities between various learning rules that might be employed by ANN data miners. Since different learning rules lead to different connection weights and stability coefficients, a graphical representation of the data that provides a novel visual means of discerning these similarities and differences is demonstrated. The second part of this research demonstrates a methodology for ANN model variable interpretation that uses network connection weights. It uses empirical marketing data to optimize an ANN and response elasticity graphs are built for each ANN model variable by plotting the derivative of the network output with respect to each variable, while changing network input in equal increments across the range of inputs for each variable. Finally, this paper concludes that such an approach to ANN model interpretation can provide data miners with a rich interpretation of variable importance.

Citation

Fish, K.E. and Segall, R.S. (2004), "A visual analysis of learning rule effects and variable importance for neural networks in data mining operations", Kybernetes, Vol. 33 No. 7, pp. 1127-1142. https://doi.org/10.1108/03684920410534461