Modelling cellphone trace travel mode with neural networks using transit smartcard and home interview survey data

  • James Vaughan University of Toronto Transportation Research Institute
  • Ahmadreza Faghih Imani Centre for Transport Studies, Imperial College London
  • Bilal Yusuf Metrolinx
  • Eric J. Miller University of Toronto Transportation Research Institute

Abstract

This study proposes a framework to impute travel mode for trips identified from cellphone traces by developing a deep neural network model. In our framework, we use the trips from a home interview survey and transit smartcard data, for which the travel mode is known, to create a set of artificial pseudo-cellphone traces. The generated artificial pseudo-cellphone traces with known mode are then used to train a deep neural network classifier. We further apply the trained model to infer travel modes for the cellphone traces from cellular network data. The empirical case study region is Montevideo, Uruguay, where high-quality data are available for all three types of data used in the analysis: a large dataset of cellphone traces, a large dataset of public transit smartcard transactions, and a small household travel survey. The results can be used to create an enhanced representation of origin-destination trip-making in the region by time of day and travel mode.

How to Cite
VAUGHAN, James et al. Modelling cellphone trace travel mode with neural networks using transit smartcard and home interview survey data. European Journal of Transport and Infrastructure Research, [S.l.], v. 20, n. 4, p. 269-285, oct. 2020. ISSN 1567-7141. Available at: <https://journals.open.tudelft.nl/ejtir/article/view/5429>. Date accessed: 26 jan. 2021. doi: https://doi.org/10.18757/ejtir.2020.20.4.5429.
Published
2020-10-01