From Data to Decisions: A Comparative Study of Machine Learning and Statistical Approaches in Estimating Passenger Counts.
Year: 2024
Type: dissertation
Abstract: Urban transport planning is inherently complex, requiring accurate representation of transport demand and mobility patterns across spatial and temporal dimensions.Intelligent Transport Systems (ITS) have emerged as pivotal in enhancing the efficiency and environmental sustainability of urban transport networks by automating data collection and processing.Modern technologies, such as mobile applications, Geographic Information Systems (GIS), Internet of Things (IoT) devices, and 5G connectivity, have significantly improved data quality and quantity, outperforming traditional methods like telephone surveys.The transport sector has used data collection methods such as passenger counting (APC) systems to estimate mobility patterns and improve public transport services.Numerous studies in literature have sought to refine the data collected through these systems, aiming to provide a more precise understanding of mobility trends.While existing literature has focused on enhancing univariate data for prediction and categorisation, efforts have also been made to integrate external variables like environmental and meteorological data to improve prediction accuracy.However, despite these advancements, selecting appropriate systems or processes remains challenging due to the lack of standardised protocols, often resulting in sub-optimal implementations.Moreover, determining the best method for specific datasets is still difficult, as few studies have compared these models using real operational data or considered factors like dataset size and computational power.This research addresses these gaps by empirically evaluating passenger count estimation models using both artificial intelligence (AI) and traditional statistical techniques, with a focus on data fusion, using real operation data.It also examines the discrepancies between claimed and actual performance of Automated Passenger Counting (APC) systems, which often report accuracy rates of 98% to 99%, yet real-world performance varies widely between 50% and 78%.Specific objectives include developing and implementing diverse passenger count estimation models, applying methods from other fields to the transport sector, evaluating model performance through comprehensive metrics, and providing a comparative analysis to identify each model's strengths and weaknesses in passenger count estimation.Words cannot express my gratitude to my supervisor Prof. Cristina Pronello for her invaluable supervision, support and tutelage during the course of my PhD degree.Special thanks are extended to Deepan Anbarasan and Carlos Garcia for their technical support and proofreading my study, and to the TRIS research group, particularly Dr. Pinky Kumawat and Daniel Sebastia, for their immense support during the data collection phase.Their help made it possible to endure the long hours spent on the bus counting passengers Additionally, this endeavour would not have been possible without the support of Leonardo S.p.A. who financed my research, and to GTT and Mobyforall for their collaboration and for enabling me to work with cutting-edge technologies on their vehicles.Finally, I would like to express my gratitude to my parents, my sisters
Source: PORTO@Iris
Author: GARZON RUIZ, XIMENA ROCIO
Language: English
Subfield: Transportation
Field: Social Sciences
Domain: Social Sciences
Keywords: machine learning, statistical analysis, model comparison, data fusion, public transport demand, urban transportation planning, intelligent transportation systems, model evaluation, meteorological observations, Internet of Things, 5G connectivity, computational power, dataset size, model interpretability, explainable AI
Open Access status: green
OpenAlex ID: W7217796632
