Optimising public transit using big data and machine learning

• Main Author: Lee, Kelvin
• Other Authors: Su Rong
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering
• Online Access: https://hdl.handle.net/10356/173342
• Institution: Nanyang Technological University

Despite decades of research on optimisation of public transit, recent advances in big data collection and machine learning methods have created new possibilities for further optimisation. In this thesis, the main objective is to optimise existing public transit services by utilising big data without provision for extra resources. We propose innovative methods and examine the extent to which public transit can be improved through data-driven optimisation alone. In the first part of the thesis, we explore a predictive optimisation framework that employs machine learning techniques to forecast passenger demand, which are then used to optimise a demand-responsive transit service. Our findings demonstrate that using the predictive optimisation framework results in superior and more resilient solutions than relying on point estimates. We also characterised the relationship between prediction accuracy and the quality of solutions from downstream optimisation tasks. In the second part of the thesis, we investigate the potential of leveraging large amounts of high-resolution data collected by transit operators to improve public transit services. We investigate how utilising travel time and demand data obtained at higher frequencies in optimisation models can lead to better synchronisation and thus passenger experience. Furthermore, we explore the potential of opportunistically operating limited-stop services alongside existing all-stop services, based on observed travel times and demand to reduce passenger travel time. In conclusion, this thesis introduces multiple techniques for transit operators to utilise their existing data to conduct data-driven optimisation of their services. Our findings indicate that with careful planning and inventive methods, big data and machine learning can aid public transit operators in running their services more efficiently, thereby making them more attractive to passengers.