Skip to Main content Skip to Navigation

Estimating Urban Mobility with Mobile Network Geolocation Data Mining

Abstract : In the upcoming decades, traffic and travel times are expected to skyrocket, following tremendous population growth in urban territories. The increasing congestion on transport networks threatens cities efficiency at several levels such as citizens well-being, health, economy, tourism and pollution. Thus, local and national authorities are urged to promote urban planning innovation by adopting supportive policies leading to effective and radical measures. Prior to decision making processes, it is crucial to estimate, analyze and understand daily urban mobility. Traditionally, the information on population movements has been gathered through national and local reports such as census and surveys. Still, such materials are constrained by their important cost, inducing extremely low-update frequency and lack of temporal variability. On the meantime, information and communications technologies are providing an unprecedented quantity of up-to-date mobility data, across all categories of population. In particular, most individuals carry their mobile phone everywhere through their daily trips and activities. In this thesis, we estimate urban mobility by mining mobile network data, which are collected in real-time by mobile phone providers at no extra-cost. Processing the raw data is non-trivial as one must deal with temporal sparsity, coarse spatial precision and complex spatial noise. The thesis addresses two problematics through a weakly supervised learning scheme (i.e., using few labeled data) combining several mobility data sources. First, we estimate population densities and number of visitors over time, at fine spatio-temporal resolutions. Second, we derive Origin-Destination matrices representing total travel flows over time, per transport modes. All estimates are exhaustively validated against external mobility data, with high correlations and small errors. Overall, the proposed models are robust to noise and sparse data yet the performance highly depends on the choice of the spatial resolution. In addition, reaching optimal model performance requires extra-calibration specific to the case study region and to the transportation mode. This step is necessary to account for the bias induced by the joined effect of heterogeneous urban density and user behavior. Our work is the first successful attempt to characterize total road and rail passenger flows over time, at the intra-region level. Although additional in-depth validation is required to strengthen this statement, our findings highlight the huge potential of mobile network data mining for urban planning applications.
Complete list of metadata

Cited literature [106 references]  Display  Hide  Download
Contributor : Danya Bachir Connect in order to contact the contributor
Submitted on : Thursday, February 21, 2019 - 10:55:52 AM
Last modification on : Monday, December 13, 2021 - 9:16:37 AM
Long-term archiving on: : Wednesday, May 22, 2019 - 3:02:56 PM


Files produced by the author(s)


  • HAL Id : tel-02043700, version 1


Danya Bachir. Estimating Urban Mobility with Mobile Network Geolocation Data Mining. Data Analysis, Statistics and Probability []. Télécom SudParis (Institut Mines-Télécom); Université Paris Saclay, 2019. English. ⟨tel-02043700⟩



Record views


Files downloads