Institute of Transport and Logistics Studies

Analysis of travel behaviour using artificial neural networks: uncovering individuals’ value of time

Dr Sander van Cranenburgh, Delft University of Technology, The Netherlands

17th May 2018  02:00 pm - 03:00 pm ABS SR3090, Level 3, Abercrombie Building (H70)

Abstract: Recently, Artificial Neural Networks (ANNs) are gaining popularity in many research fields. ANNs are mathematical models which are loosely inspired by the structure and functional aspects of biological neural systems. However, despite the general excitement about the potential of ANNs (and other data-oriented techniques), their use within the travel behaviour research field is still fairly limited. This limited use is mainly due to the back-box nature of ANNs. In this seminar I will show that the black-box nature of ANNs does not preclude developing new insights on travel behaviour using ANNs. In particular, this study develops a new ANN-based methodology to uncover heterogeneity in the Value-of-Time (VoT). Using this methodology it is possible to accurately pinpoint each individual’s VoT based on a series of observed choices (which embed trade-offs between travel cost and travel time). This line of research may open up all sorts of new opportunities, leading to a richer understanding of individual level behaviour in transport, as well as beyond.

Biography: Sander van Cranenburgh is assistant Professor in the Transport and Logistics (TLO) group of Delft University of Technology. Sander his research focusses on choice behaviour analysis. During his Postdoc he mainly worked on Random Regret Minimization (RRM) based discrete choice models, and made several important methodological contributions to this modelling approach. His current research increasingly focusses on enriching understanding of (travel) choice behaviour through developing new data-driven modelling approaches. In this research he specifically seeks the edge between traditional theory-driven approaches, such as Discrete Choice Models, and data-driven approaches, such as Artificial Neural Networks.

For further details please contact