Inferring dynamic origin-destination flows by transport mode using mobile phone data

Danya Bachir (1, 2, 3, 4, 5) , Ghazaleh Khodabandelou (1, 2, 3, 6, 4) , Vincent Gauthier (1, 2, 3, 4) , Jakob Puchinger (7, 5) , Mounim El Yacoubi (1, 8, 9)

1 IP Paris - Institut Polytechnique de Paris
2 TSP - RST - Département Réseaux et Services de Télécommunications
3 NeSS-SAMOVAR - Network Systems and Services
4 CNRS - Centre National de la Recherche Scientifique
5 IRT SystemX
6 LPTMC - Laboratoire de Physique Théorique de la Matière Condensée
7 LGI - Laboratoire Génie Industriel - EA 2606
8 TSP - EPH - Département Electronique et Physique
9 ARMEDIA-SAMOVAR - ARMEDIA

Fast urbanization generates increasing amounts of travel flows, urging the need for efficient transport planning policies. In parallel, mobile phone data have emerged as the largest mobility data source, but are not yet integrated to transport planning models. Currently, transport authorities are lacking a global picture of daily passenger flows on multimodal transport networks. In this work, we propose the first methodology to infer dynamic Origin-Destination flows by transport modes using mobile network data e.g., Call Detail Records. For this study, we pre-process 360 million trajectories for more than 2 million devices from the Greater Paris as our case study region. The model combines mobile network geolocation with transport network geospatial data, travel survey, census and travel card data. The transport modes of mobile network trajectories are identified through a two-steps semi-supervised learning algorithm. The later involves clustering of mobile network areas and Bayesian inference to generate transport probabilities for trajectories. After attributing the mode with highest probability to each trajectory, we construct Origin-Destination matrices by transport mode. Flows are up-scaled to the total population using state-of-the-art expansion factors. The model generates time variant road and rail passenger flows for the complete region. From our results, we observe different mobility patterns for road and rail modes and between Paris and its suburbs. The resulting transport flows are extensively validated against the travel survey and the travel card data for different spatial scales

Mots clés: Origin destination matrix; Mobile phone data; Transport mode; Machine learning; Travel flows; Urban mobility

Domaines: Intelligence artificielle [cs.AI]; Réseaux et télécommunications [cs.NI]; Analyse de données, Statistiques et Probabilités [physics.data-an]; Génie civil; Machine Learning [stat.ML]

Danya Bachir, Ghazaleh Khodabandelou, Vincent Gauthier, Jakob Puchinger, Mounim El Yacoubi. Inferring dynamic origin-destination flows by transport mode using mobile phone data. Transportation research. Part C, Emerging technologies, 2019, 101, pp.254-275. ⟨10.1016/j.trc.2019.02.013⟩. ⟨hal-02043639⟩