Modeling railway passenger overall comfort: combining PLS-SEM and interpretable machine learning

Based on Stimuli-Organism-Response framework, this study examines how carriage environmental quality perception (CEQP), valence, exposure time, and individual factors affect passenger comfort. Through field investigation, PLS-SEM combined with machine learning was employed to identify and quantify the contribution of factors affecting passenger overall comfort. The results indicate that CEQP, valence, exposure time, age, agreeableness, and environmental sensitivity significantly influence overall comfort. Valence mediates the relationship between CEQP and overall comfort. Pressure quality perception (PQP) has the greatest impact on overall comfort. Significant factors were incorporated into the Adaptive Boosting (ADA) model. ADA-SHAP analysis revealed that PQP made the largest contribution (27.08%) to overall comfort, exceeding other environmental dimensions, followed by agreeableness (20.37%), valence (15.10%) and exposure time (11.47%). Other influencing factors also contributed to the model to some extent. These findings provide guidance for environmental regulation, route selection, and comfort optimization of train carriages