Evolution of urban travel mode choice modelling from discrete choice models to machine learning

This paper presents a critical literature review of how travel mode choice modelling has evolved, particularly concerning the shift towards machine learning (ML), deep learning (DL), and explainable artificial intelligence (XAI) frameworks. The systematic literature search was done through the Scopus database, and the inclusion criteria and screening procedures have been defined to identify the relevant studies published between 1998 and 2026. Moreover, bibliometric analysis methods were also used to investigate the overall trend of research, collaboration patterns, and topics of development in the area. The results demonstrate a definite change in the methodology due to the growing availability of data and the necessity to model non-linear and complex travel behaviour. Although discrete choice models (DCMs) are necessary to interpret behaviours and analyse policies, ML and DL models show a high predictive accuracy, especially in data-abundant settings. Nevertheless, they can only be applied in the policy contexts to a limited extent due to their lack of interpretability. XAI techniques and hybrid modelling approaches have become an attractive option to address this gap by combining predictive performance with behavioural transparency. The review also reveals major research gaps, such as the necessity of causal inference in data-driven models, better model transferability, and the successful combination of heterogeneous data sources. Such understandings form the basis of the next-generation mode choice models to aid in both precise prediction and policy-appropriate decision-making in transportation planning.