Towards eXplainable AI for Mobility Data Science

Temporal graph neural networks can make mobility predictions more interpretable: The paper proposes using temporal GNNs to analyze dense trajectory data (GPS tracks from vehicles and vessels) in ways that can be explained to humans, rather than treating these systems as "black boxes."

Counterfactual explanations offer practical insights for mobility applications: By showing "what-if" scenarios (e.g., "if the vehicle had taken a different route"), counterfactual approaches help stakeholders understand why a model made specific predictions about movement patterns.

Human-centered design is essential for explainable AI in mobility: The authors emphasize that explanations must be comprehensible to actual users and decision-makers, not just technically sound—a critical consideration for applications affecting transportation, logistics, and urban planning.

Current geographic AI research lacks sufficient focus on explainability: The paper reviews existing GeoXAI (geographic explainable AI) studies and identifies a gap in the field, suggesting that mobility data science has lagged behind other domains in developing transparent, understandable models.

Explainable mobility AI has real-world governance implications: Making AI systems that track and predict movement patterns more transparent is important for responsible deployment in contexts involving privacy, public safety, and infrastructure decisions.