Comparison of Logit and Neural Network Models in Inter-Island Discrete Choice Analysis

Authors: Krister Ian Daniel Roquel and Alexis Fillone

Abstract

Logit-based models have often been used for discrete choice analysis. However, conventional logit models preserve a linear relationship that requires variables that are independent of each other, which is generally not the proper assumption. In this paper, the researcher addresses the non-linear behavior and inter dependence of variables using neural networks in modeling inter-island travel choice. Neural network analysis was employed to a previous work to test the applicability of neural network in discrete choice models for inter-island travel. It was found that the neural network model is statistically acceptable in describing travel choice behavior, while the logit model is more inclined to model the decision making process. Also, it was found that the neural network model is capable of accurately predicting the minority, which has long been a problem when using logit models as these are usually treated as errors.