Artificial Neural Networks and Discrete Choice Models: Comparing and Contrasting

Artificial neural networks and discrete choice models have been successfully applied to the prediction of individual choices. This research compares the different models to determine which has better predictive ability. From the discrete choice models, the multinomial logit model and the mixed logit model have been selected. Among the Artificial Neural Networks, Deep Neural Networks, Genetic Artificial Neural Networks, different variants of Backpropagation Algorithm, Radial Basis Functions, and Constructive Algorithms were selected. All the models were evaluated on three datasets related to product purchases in supermarkets. For each dataset, all models were run 30 times. The percentage of correctly classified observations was used to evaluate the performance. Results showed that Deep Neural Networks, Genetic Artificial Neural Networks, Levenberg–Marquardt Backpropagation, Resilient Propagation, and Probabilistic Neural Networks outperformed discrete choice models. However, discrete choice models are superior in terms of robustness. They require fewer resources and provide additional information to the decision-maker.