<p>Encouraging cycling is an important strategy for promoting sustainable urban transportation. This study examines how environmental characteristics along cycling routes are associated with cyclists’ decisions to deviate from navigation-recommended routes by comparing actual cycling trajectories with navigation-recommended routes. Using large-scale shared bicycle GPS data from Xiamen, China, the analysis integrates map-matched cycling routes with recommended alternatives generated by a navigation platform. An XGBoost-based classification model is employed together with SHAP and GeoSHAP methods to explore the relative importance, nonlinear patterns, and spatial variability of route-level environmental factors. The results indicate that cyclists’ route deviations are closely associated with multiple environmental attributes. Cycling infrastructure quality and street-level green visibility are generally linked to higher preference for actual routes, while higher road hierarchy, greater commercial mixing, and increased intersection density tend to be associated with lower route attractiveness. Several factors exhibit clear nonlinear relationships, suggesting that cyclists’ responses vary across different environmental conditions rather than following simple linear patterns. Spatial analysis further reveals notable heterogeneity in these associations. The influence of cycling infrastructure is more pronounced in areas with fragmented networks, whereas green visibility shows stronger associations in peripheral districts. In contrast, commercial mixing exhibits more evident effects in dense urban cores. Overall, the findings highlight the context-dependent nature of cyclists’ route deviation behavior and provide evidence to support place-sensitive cycling environment planning.</p>