ABSTRACT
Conventional disaggregate choice models consider choice to be an outcome of individual preferences. A variety of estimating procedures have been developed to calibrate these models and can be considered “mapping mechanisms” which link a dependent variable (e.g., choice of destination) with independent variables (e.g., individual characteristics such as age, gender, and income and destination-related attributes including attractiveness and cost). A new approach has been developed to achieve this “mapping” and is generally referred to as “artificial neural networks” (ANNs).
The goal of this study was to evaluate the usefulness of artificial neural networks for modeling individual choice behavior. The two specific objectives of this study were to compare the accuracy in prediction of a conventional conjoint model and an ANN model and to compare part-worth utilities or preference structure of the respective models. Comparison between a conventional disaggregate choice model (e.g., the conjoint model) and a back-propagation neural network (BPNN) model was based upon a study involving on-site interviews of visitors to five Illinois Highway Welcome Centers.
The findings of the study indicate that the back-propagation neural network approach is a useful approximator or pattern matcher for discrete choice models. The BPNN model performs equally well or 120better than the more traditional conjoint model in terms of goodness-of-fit and prediction rate. However, the findings of BPNN analysis also exhibited contradictory results to conjoint analysis. Two out of six models calibrated provided totally different results. Implications of the findings for travel and tourism marketing are discussed.
