Research Article
Machine Learning Methods to Forecast Public Transport Demand Based on Smart Card Validations
@INPROCEEDINGS{10.1007/978-3-031-49379-9_11, author={Brunella Caroleo and Silvia Chiusano and Elena Daraio and Andrea Avignone and Eleonora Gastaldi and Mauro Paoletti and Maurizio Arnone}, title={Machine Learning Methods to Forecast Public Transport Demand Based on Smart Card Validations}, proceedings={Intelligent Transport Systems. 7th EAI International Conference, INTSYS 2023, Molde, Norway, September 6-7, 2023, Proceedings}, proceedings_a={INTSYS}, year={2023}, month={12}, keywords={public transport demand machine learning clustering forecasting}, doi={10.1007/978-3-031-49379-9_11} }- Brunella Caroleo
Silvia Chiusano
Elena Daraio
Andrea Avignone
Eleonora Gastaldi
Mauro Paoletti
Maurizio Arnone
Year: 2023
Machine Learning Methods to Forecast Public Transport Demand Based on Smart Card Validations
INTSYS
Springer
DOI: 10.1007/978-3-031-49379-9_11
Abstract
This paper explores the forecasting of public transport demand using mobility data obtained from electronic tickets and smart cards. The research aims to estimate the demand for a selected route at a specific bus stop on a given day and time slot. The study utilizes a large dataset of historical demand data, including approximately 10 million validations collected in 2019 by the Piedmont transport operator Granda Bus, and combines it with additional information such as weather conditions, anonymized user data, and temporal segmentation of the yearly calendar. To identify the peculiarities in demand forecasting for each bus route and stop, a clustering analysis is performed, resulting in the identification of six cohesive and homogeneous clusters. Various machine learning models are tested and compared to determine the most suitable model for forecasting public transport demand at each stop within one-hour time slots. The results demonstrate that machine learning algorithms consistently outperform average-based techniques: the machine learning algorithms exhibit a significant improvement (up to 50% compared to the baseline) when demand uncertainty is greater. The proposed methodology framework is replicable and transferable to other areas, providing a valuable tool for optimizing resource allocation and network planning, while enhancing user satisfaction by accurately forecasting passenger demand at each stop and desired time slot.
