Flexible model structure for discrete-choice models

Discrete choice models are widely used for travel demand analyses, and several such models were developed based on random utility theory. The estimation process in such models generally uses the entire study population and the same model structure to estimate average parameter values, but this approach limits the possiblility that the model can fully explain the behaviour of different populations. A new approach, the flexible model structure (FMS), expands existing discrete choice models with the addition of two main components: the segmentation process and model structure search. The aim of the FMS process is to guide the progress of model estimation towards a more behaviourally realistic representation. The main contribution of this paper is to suggest and illustrate a framework that simultaneously searches for the best segmentation and the best model structure for each segment. This paper presents a numerical case study that illustrates the FMS concept. The results indicate that (a) the addition of the segmentation process acts to emphasise the heterogeneity that exists among segments and demonstrates the importance of segmentation by significantly improving the estimation results; and (b) the model structure can vary along segments.