ABSTRACT
Artificial neural networks (ANN) have become increasingly popular. Modelled loosely after the human brain, ANNs are intelligent, in the way they can learn and generalize by example. This chapter helps dispel the belief that the neural network is a ‘black box’. Specifically, it demonstrates a unique connection between the multilayer feedforward neural network model and the binary logit model, where the former may be classified as a logit model without the restriction of linearity in the parameters in the utility function. The chapter provides the concepts of adaptive logit model, adaptive regression coefficients for the utility function, and adaptive direct and cross-elasticities of demand. It shows how through a special design of the neural network, and through the use of backpropagation as a training algorithm, one can derive the adaptive regression coefficients and elasticities of demand.
