.png){kind=icon}
.png){kind=icon}
Return
A MACHINE LEARNING APPROACH TO DETERMINING FACTORS INFLUENCING PASSENGER RAIL TRAFFIC
.png){kind=icon}
PRE
AI
B
K
DOI：10.17270/J.LOG.001398.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Background: Passenger rail demand forecasting is essential for designing efficient and sustainable transport systems. This study develops a machine-learning framework to forecast both regional and long-distance passenger rail workloads in Poland, based on representative lines from each voivodeship. The modelling pipeline integrates ensemble tree-based algorithms, evolutionary feature selection, hyperparameter optimization, and post-hoc explainability using SHAP to ensure both predictive accuracy and interpretability. Methods: The framework combines ensemble machine learning models with evolutionary feature selection and hyperparameter optimization. Model interpretability and the contribution of individual variables were examined using SHAP analysis. Results: Random Forest consistently provides robust baseline performance, while optimally tuned boosting algorithms, particularly XGBoost, achieve the highest predictive accuracy, with a mean sMAPE of 0.595. Evolutionary feature selection reduced dimensionality while improving model accuracy, indicating that a smaller subset of variables captures the key determinants of passenger demand. SHAP analysis shows that regional demand is primarily driven by population size, accommodation capacity, and commuting intensity, with multimodal public transport connections amplifying these effects. In contrast, long-distance demand is mainly shaped by commuter flows, spatial accessibility, and feeder networks, while population size plays a secondary role. Conclusions: The proposed framework contributes both methodological and practical insights. Methodologically, it demonstrates how ensemble models can be combined with evolutionary feature selection and interpretability techniques to produce transparent demand forecasts. Practically, the results provide guidance for transport planning: regional demand is strongly influenced by population (21.7%), the number of small service units (16.9%), and tourism-related factors ( 16.5%), supported by multimodal integration and local commuting flows. Long-distance demand is primarily shaped by commuting intensity (28.5%), population (16.3%), and the transport accessibility of railway stations and stops (15.5%), measured by the number of bus and tram connections serving them. These findings support evidence-based planning of passenger rail services in Poland and may also be relevant for rail transport planning in Central and Eastern Europe.
Keywords:
rail transport
passenger demand forecasting
machine learning
ensemble models
evolutionary feature selection
SHAP interpretability
Journal
L
IF:
1
Papers:
23
Citations:
402
Organization
Cited Papers