With mobile phone technologies continuing to spread around the world, Call Detail Records (CDR) represent an attractive additional data source for inferring internal migration in addition to traditional data sources such as surveys and administrative records, especially in low and middle-income countries (LMICs), where the traditional data sources can be either unavailable or difficult to collect. Stay location is the basis of the most mobility statistics derived from location data and CDR. Detecting stay location is crucial to migration statistics (Where do people live? When and to where they change their residence?), to disaster statistics (Where have people lost their homes? Where are they displaced to? Did they then return home and when?), and to a multitude of other applications from informing disease spread to tourism statistics. Though a number of methods were proposed in the literature to detect stay locations from CDR data, some challenges remain especially in LMICs - where CDRs tend to be sparser - such as irregularity and low frequency of phone use, network instability, so called ‘ping-pong’ and ‘teleportation’ effects as artefacts of mobile communications, and resulting confusion between changes in phone usage and changes in mobility.
In this paper, we present our solution to detect stay locations (long and short stays) and relocations from CDRs, which addresses the above issues related to particularly sparse data in LMICs and can be run on mobile operator infrastructure (constrained in memory and compute power) to ensure data privacy.
Read the full abstract
Flowminder Foundation, Veres G., et al. (2023). Correcting measurement biases in the detection of long and short stay locations in sparse Call Detail Records (CDRs). https://doi.org/10.5281/zenodo.8414389