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

Abstract

Green mobility solutions are increasingly integrated into public spaces to enhance accessibility and promote sustainable tourism. This study examines the role of green mobility vehicles—small, electric assistive transport units—in influencing visitor spatial behavior within Wakayama Castle Park, a large historical site in Japan. Using a structured questionnaire dataset that captures both intended destinations (without mobility support) and actual destinations (after using mobility vehicles), we develop a binary classification model to predict whether the use of green mobility vehicles encourages visitors to reach farther points in the park. The model incorporates features such as demographic profiles, spatial distance measures, and visit characteristics. An XGBoost classifier achieves a 77% accuracy rate, demonstrating the significance of green mobility as a factor in spatial expansion. To support interpretability and policy insight, we integrate DiCE (Diverse Counterfactual Explanations), enabling post-hoc exploration of how slight changes in inputs (e.g., age, trip purpose) can influence outcomes. A custom interface is developed to connect DiCE with the XGBoost model. Our results suggest that green mobility vehicles can significantly alter visitor movement patterns, and Combining predictive and counterfactual modeling provides a practical approach to evaluate their spatial and behavioral impact in urban recreational environments.