查看文章

Discrete choice models have been an integral part of transport research including but not limited to road safety analysis, transport planning, land-use, and traffic operations. Even with the advent of advanced machine learning (ML) methods, discrete choice models are still widely used because of their ability to capture behavior and estimate causality. Over the years, extensions and capabilities have been added to better capture behavior, but each with their own strengths and limitations (Mannering et al., 2016).

The process of developing discrete choice specifications is highly involved as it requires the analyst to take critical modelling decisions, including (1) selection of variables to be tested during model specification;(2) identification of variable forms and transformations (eg, linear or non-linear);(3) variables to be tested with fixed coefficients;(4) variables to be tested with random coefficients;(5) distributional assumptions for the random coefficients and error terms;(6) selection of methods to deal with potential correlation.