Abstract
The rapid development in telecommunication networks is producing a huge amount of information regarding how people (with their mobile devices) move and behave over space and time. While GPS data, typically collected by smartphone apps, are restricted to rather small samples of the population, mobile phone network data, routinely collected by mobile network operators, potentially allow to analyze travel behaviors and social interaction of the whole population, with full temporal (e.g., longitudinal) coverage at a comparatively low cost. Therefore, recent years have seen an increasing interest in using such data for human mobility studies. However, due to their noisy and temporally infrequent/irregular nature, extracting mobility information such as transport modes from these data is particularly challenging. This paper provides an in-depth, systematic review of transport mode detection based on mobile phone network data. The results of the review show that existing studies tend to identify easy-to-detect modes (e.g., train or metro), or aggregate fine-grained modes into more general groups (e.g., public versus private transport). Rule-based methods making use of geographic data were often developed. More importantly, due to the lack of ground truth data, evaluation of the proposed methods was seldom done and reported. Finally, we identify a list of research gaps currently being faced in this field, particularly regarding robust and real-time data cleaning and mode detection methods, “benchmark” datasets and metrics allowing the comparison of different mode detection methods, as well as privacy and bias issues.