Analysis and Prediction of the Commuting Mode Choices in England Using Bayesian Networks

Understanding commuting travel behaviour is crucial for addressing significant issues in sustainability and development. This research aims to develop a novel Bayesian Network (BN) model that enables rapid investigations of complex influences upon travel mode choices through standard travel, place, and work surveys, which would become a new method that complements the existing models. The theoretical framework underpinning BNs, particularly those involving Directed Acyclic Graphs (DAGs), is recently attracting new attention by offering an intuitive and visual representation of causal relationships that align more closely with human reasoning than traditional statistical models. Among existing travel mode choice models, discrete choice models based on random utility theories have been widely used – they are effective in practical use, particularly when travel survey data is supplemented with information on non-chosen modes. However, the formulation of their utility functions makes it hard to consider a complex web of direct and indirect influences simultaneously and intuitively and would require technical skills not available in most cities and transport agencies. Structural Equation Models (SEM), on the other hand, combines factor and path analysis to understand the key direct and indirect influences in an intuitive way, but it is usually severely limited by data size and the number of influences that can be tested simultaneously. Other types of machine learning models have been developed but they generally lack clear and transparent causal representation. In these respects, a Bayesian Network (BN) model offers a promising alternative for investigating a more complex web of influences. Such models are already used extensively in medical research. In urban studies, BN models haven been recently applied in producing synthetic populations, for example, for urban “digital twins”. However, in spite of their potential, BNs are rarely developed for modelling travel mode choice or travel behaviour more generally. This research develops a new BN model for commuting mode choices in England, using data from the National Travel Survey from 2011 to 2019, and the 2011 Population Census microdata. The Census Microdata is used as a reference for variable selection and discretization of NTS data. The granularity of the Census Microdata also complements the NTS data. Both good quality datasets have enabled a comprehensive investigation of the properties of the BN model. The contribution of the proposed Bayesian Network model is fivefold. Firstly, this study distinguishes all main modes of travel. This level of detail represents a significant improvement over existing models that aggregate public transport and active modes. Secondly, the existing BN models struggle with accurately modelling underrepresented modes, especially biking. This research overcomes this limitation by proposing a BN model that performs well for this mode. Thirdly, the BN model incorporates socio-demographic and workplace variables in a flexible manner, allowing for an exploration of both direct and indirect influence on travel mode choices. Fourthly, this study addresses the concern over the stability of BN structures, adding reliability to the findings. The model’s stability is validated using bootstrap resampling techniques, different data sources, and temporal segmentation of the National Travel Survey dataset, ensuring the reliability of the proposed network structure. This may also shed light on potential improvements in population synthesis BN models. Lastly, the methodology of this research provides a replicable framework to adapt BN models to other datasets and regions. The research is focused on commuting, but the method is extendable to other travel purposes given suitable data. Once the BN model is established, inferences are made which show new policy implications regarding the effectiveness of interventions tailored to different population segments, such as by gender and income levels as well as across built forms. The results particularly show the importance of general travel habits on commuting mode choices.

Name dc:type.qualificationname

Doctor of Philosophy (PhD)

Level dc:type.qualificationlevel

Doctoral

Grantor dc:publisher.institution

University of Cambridge

Year dc:date.issued

2024

Author dc:creator

• Pan, Lingzi

Advisor dc:contributor.advisor

• Jin, Ying

• urban modelling
• Bayesian Networks
• transport mode choice modelling

Licence

• https://www.repository.cam.ac.uk/bitstreams/e1c39ec2-92ef-4352-a116-999cff387388/download
• http://purl.org/NET/rdflicense/allrightsreserved

Language dc:language

eng

DOI dc:identifier.doi

https://doi.org/10.17863/CAM.124589

OAI identifier oai:identifier

oai:www.repository.cam.ac.uk:1810/394832