A Questionnaire-Only Counterfactual Machine Learning Approach to Assess the Spatial Impact of Green Mobility Vehicles in Urban Parks: A Wakayama Castle Case Study

Research Note

In a large historical park, slopes and distances keep some visitors from reaching places they would like to see. Small electric assistive vehicles relax that constraint, but confirming that they actually change where people go would require comparing the same visitor with and without the vehicle, and only one of those outcomes can ever be observed.

Counterfactual reasoning is the way this unobserved side is handled. By estimating from a model what would have happened under conditions that did not occur, it allows the effect of an intervention to be assessed causally rather than merely correlationally.

This study of Wakayama Castle Park uses a questionnaire designed to capture both sides directly, asking respondents where they had intended to go without mobility support and where they actually went after using a vehicle. A binary classifier predicts whether use of the vehicles encourages visitors to reach farther points, drawing on demographic profiles, spatial distance measures and visit characteristics. An XGBoost classifier reached 77% accuracy.

DiCE, a method that generates diverse counterfactual explanations showing how inputs would need to change to flip a prediction, was then integrated through a purpose-built interface so that the influence of small changes in age or trip purpose could be explored after the fact. Combining predictive and counterfactual modelling gives a practical way to evaluate the spatial and behavioural impact of such vehicles.