Evaluating Machine Learning and Discrete Choice Models for Travel Mode Choice in Air–Rail Integration

Purpose: The growing demand for sustainable and seamless transportation has positioned air–rail integration as a key component of future mobility systems. Traditional discrete choice models, while widely used for travel mode analysis, are limited in capturing complex, non-linear passenger behaviour. In contrast, machine learning (ML) models offer strong predictive capabilities but lack sufficient behavioural validation in air–rail contexts. This study aims to develop a novel evaluation framework for analysing travel mode choice in air–rail integration while applying an established experimental methodology to assess ML models.
Research Approach: The study adopts an existing experimental methodology that structures model evaluation around synthetic datasets and controlled experiments. Within this approach, three experiments are conducted: (1) evaluating predictive accuracy under varying utility specifications and error structures; (2) assessing the calibration and behavioural consistency of predicted probabilities; and (3) analysing model performance in estimating market shares and willingness-to-pay (WTP) for integrated air–rail options, including hypothetical scenarios. Multiple ML and deep learning models are compared with the Multinomial Logit model. The synthetic data are newly designed for the air–rail context to approximate realistic passenger decision-making in integrated air–rail systems.

Findings and Originality: As the study is ongoing, planned findings will compare models in terms of predictive performance, behavioural consistency, and policy relevance. The originality lies in generating and using air–rail-specific synthetic data to transfer and test an established methodology within an underexplored multimodal context.

Research Impact: The study will show how ML-based travel mode choice can be operationalised in the air–rail systems, strengthening the link between predictive modelling and behavioural interpretation for integrated transport systems and providing a basis for future framework development focused on air–rail evaluation.

Practical Impact: The research supports transport planners and policymakers by identifying modelling approaches that are both accurate and behaviourally meaningful, enabling improved demand forecasting, pricing strategies, and service design in air–rail systems.