Psychological and built environment influences on bridging mode choices during metro service disruptions: a hybrid SEM-ANN approach

This study investigates passenger bridging-mode choices during metro disruptions by integrating psychological factors and station-area built environments. Using 1,875 valid responses from Fuzhou Metro, we combine structural equation modeling for theory-based screening, artificial neural networks for nonlinear estimation, supervised tree-based construction of eight travel-purpose-oriented passenger profiles, and a psychological-weight-constrained hybrid choice model (HCM) for group-specific heterogeneity. The HCM achieves a goodness-of-fit ratio of ρ2 = 0.347 and higher predictive accuracy than conventional multinomial logit and HCM models. Ablation tests demonstrate gains from global psychological weighting, group-specific heterogeneity, and multiplicative weight aggregation. Ride-hailing and walking are more strongly associated with perceived convenience and safety, whereas emergency bridging buses, regular buses, and bike-sharing are more strongly associated with attitude, subjective norm, and perceived behavioral control. Road density, plot ratio, remaining distance, and travel purpose are associated with preferences. Built-environment coefficients represent context-specific associations because indicators co-vary across three settings.