A Generalisable Data Fusion Framework to Infer Mode of Transport Using Mobile Phone Data

Cities often lack up-to-date data analytics to evaluate and implement transport planning interventions to achieve sustainability goals, as traditional data sources are expensive, infrequent, and suffer from data latency. Mobile phone data provide an inexpensive source of geospatial information to capture human mobility at unprecedented geographic and temporal granularity. This paper proposes a method to estimate updated mode of transportation usage in a city, with novel usage of mobile phone application traces to infer previously hard to detect modes, such as bikes and ride-hailing/taxi. By using data fusion and matrix factorisation, we integrate socioeconomic and demographic attributes of the local resident population into the model. We tested the method in a case study of Santiago (Chile), and found that changes from 2012 to 2020 in mode of transportation inferred by the method are coherent with expectations from domain knowledge and the literature, such as ride-hailing trips replacing mass transport.