Understanding cyclist route choice: A hierarchical adversarial inverse reinforcement learning approach to behavioral profiling

Understanding cyclists' route choice behavior is essential for optimizing urban cycling environments and enhancing green mobility. While prior studies have identified various influencing factors, systematic integration of hierarchical decision processes and cyclist heterogeneity remains limited. This study proposes a three-level global-local-action decision framework and integrates Adversarial Inverse Reinforcement Learning (AIRL) with SHapley Additive exPlanations (SHAP) to analyze shared-bike Global Positioning System (GPS) trajectories in Shenzhen. Results show that route choice is jointly driven by global path planning, such as minimizing left turns and distance, and local environmental perception, such as road hierarchy, greenery, and building density. Action level maneuver indicators show limited direct marginal attribution after global and local factors are considered. Using a dynamic attribution–clustering method, we identify four distinct trip level behavioral archetypes-Global Planners (29.9% of trajectories), Local Perceivers (8.9%), Balanced Opportunists (51.9%), and Dynamic Adapters (9.3%)-highlighting substantial behavioral heterogeneity. These findings offer new insights into cyclist decision-making and provide empirical guidance for targeted infrastructure planning and personalized navigation services.