A bike-sharing demand prediction model integrating user heterogeneity and spatiotemporal characteristics

This study proposes a bike-sharing demand prediction framework that integrates user heterogeneity with spatiotemporal features, aiming to enhance the model’s ability to capture complex behavioral differences in urban mobility. Using the New York City bike-sharing system as a case study, multi-dimensional station-level features were constructed by combining hourly trip records, weather variables, and holiday information. Through clustering analysis, user groups with distinct behavioral patterns were identified, and both global and cluster-specific prediction models were developed based on the XGBoost algorithm. To interpret the model results, the SHAP method was introduced to quantify the marginal contribution of each feature to the prediction outcomes. Empirical results indicate that the cluster-based modeling strategy achieves higher predictive accuracy in capturing local demand variations, with mean absolute error (MAE) improvements exceeding 60% for certain user groups. Moreover, significant heterogeneity was observed in the effects of weather and temporal factors across different clusters, revealing the nonlinear behavioral responses of users to external environments. The study provides a novel approach that combines interpretability and differentiated modeling capabilities for bike-sharing demand prediction, offering valuable insights for refined management and intelligent scheduling of urban mobility systems.