A Generalized Neural Logit Model for Airport and Access Mode Choice in Germany

The purpose of this paper is to present a new kind of discrete choice model called "Generalized Neural Logit Model" applied exemplarily to the case of airport and access mode choice. This approach employs neural networks to model the utility function of a discrete choice model and correlations within the alternative set and genetic algorithms to optimize the network structure. To evaluate the new approach the application case of airport and access mode choice is chosen. Benchmark for the Generalized Neural Logit Model is a nested logit approach. The estimated market segment specific airport and access mode choice models are generally applicable to any number of airports and combinations of airports and access modes. Thereby it is possible to analyse future scenarios in terms of new airport constellations and new airport access modes. To achieve this, Kohonen’s Self-Organizing-Maps are used to identify different airport clusters and assign every airport to the appropriate cluster. Although the nested logit model show a good model fit for most market segments, the Generalized Neural Logit approach produces a significant increase in model fit especially for those market segments whose nested logit model show less satisfying results.