Analysis of ride-hailing service discontinuity: Links to built environment and public transportation

The saturation of the ride-hailing market has intensified competition among drivers for ride requests, leading to a continuous decline in their satisfaction. Understanding the impact of built environment variables on drivers' access to services has become crucial. This study defines Ride-hailing Service Discontinuity (RSD) by weighting the transfer time and distance between two consecutive rides of drivers to quantify the difficulty drivers face in accessing services. This paper explores the effects of built environment variables on RSD in Hangzhou by integrating Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) clustering, Conditional Tabular Generative Adversarial Network (CTGAN) data augmentation, Light Gradient Boosting Machine (LightGBM) combined with SHapley Additive exPlanations (SHAP) explainable machine learning technique, and Stacking Ensemble Prediction. The results indicate that the impact of built environment variables on RSD exhibits regional variability and nonlinear effects. The nonlinear effects are manifested in two forms: similar nonlinear patterns with different thresholds and changes in nonlinear form. Threshold effects are widely observed in built environment variables, and some variables (e.g., Slope) exhibit consistent marginal contribution to RSD within a specific threshold range. In addition, the Stacking Ensemble Prediction is better at capturing the inherent relationships between variables and RSD, demonstrating higher prediction accuracy. In particular, when the performance of base models is moderate, the ensemble prediction performs better. This study provides a novel approach to understanding the impact of the built environment on ride-hailing from the driver's perspective.