Understanding the complex interaction of spatial and temporal factors is crucial for proposing data-driven transport policies and infrastructure planning. This study aims to model bike counts (2014 -2024) with a gradient boosting framework, specifically XGBoost, and identify how weather-related, infrastructure, and built environment variables shape cycling behavior. The dataset covers over 1.400 monitoring locations, each capturing 15-minute counts across the Greater London Area. Shapley Additive Explanations (SHAP) were applied for both global and local interpretation of model predictions.

Results show that infrastructure variables, such as road type, cycle lane design, and proximity to bike share stations, have the strongest predictive power, followed by built environment variables. Temporal variables, especially peak hours, also contribute to the model's performance. The weather had the least effect due to limited variation in the dataset. SHAP enabled interpretation by ranking variable importance, revealing spatial variations, identifying non-linear effects, and detecting threshold values. These insights informed data-driven planning recommendations at the global, region, and borough levels.

This study demonstrates the potential combining machine learning with explainable AI to generate actionable insights. The proposed method offers a transferable framework for understanding cycling behavior and supporting data-driven planning in other cities.