Explainable Modeling of Urban Bus Inter-Stop Travel Time: Independent Cross-City Replication and Data-Driven Managerial Recommendations
Explainable Modeling of Urban Bus Inter-Stop Travel Time: Independent Cross-City Replication and Data-Driven Managerial Recommendations
Parisa GhahremanShahraki1
1) Computer Engineer
IOI Code
:
AAHE-ABKHCH
Publication :
National Conference on Civil Engineering, Design and Construction Technologies, Energy, Materials and Advanced Machinery(icaeu.ir)
Abstract :
Accurate prediction of urban bus travel time between consecutive stops is important for fleet operations management, passenger information provision, and identification of network performance bottlenecks. However, many studies focus only on maximizing model accuracy and do not report, in an interpretable form for transport managers, the contribution of factors associated with longer travel times. This study presents an explainable machine-learning framework for predicting observed bus movement time between two consecutive stops and translating the results into managerial recommendations. The primary data comprised 217,545 records from the Macau bus system, and the independent data comprised 785,976 route-segment records from Astana. After auditing, cleaning, and applying a minimum route-segment frequency criterion, 217,532 and 781,001 records, respectively, remained for modelling in the two cities. In Macau, ridge regression performed best, with a mean absolute error (MAE) of 20.33 seconds and an R-squared value of 0.844. SHAP analysis showed that route-segment identifier, traffic condition, and passenger load were the most influential determinants of travel time. Across 46 comparable segments, heavy traffic increased median travel time by 18.25 seconds (34.25%). In Astana, prospective temporal evaluation showed that CatBoost was the best model, with an MAE of 39.14 seconds and an R-squared value of 0.552; origin-stop dwell time was the most important predictor, with 51.97% importance. The results indicate that managing dwell time at the origin stop, targeting interventions at congested segments, and conducting segment-level network reviews can provide a practical basis for improving bus-system performance.
Keywords : urban bus travel time prediction explainable machine learning shap catboost public transport management cross city replication