Social distancing evaluation with network analytics.

May 26, 2020

How the analysis of mobile network activity can reveal changes in social distancing patterns.

COVID-19 has caused terrible damage since it erupted in late 2019. Governments all over the globe scrambled to respond to it, often overwhelmed by the speed with which the disease spread. In face of all the confusion about what measures to adopt, perhaps one near-universal rule emerged: that social distancing had to be in place for more lives to be saved.

As a result, telecommuting, cancellation of public events, closure of shopping malls or downright curfews were swiftly instated by national and local authorities. When followed properly, those policies are supposed to make it less likely for people to physically run into each other and, ultimately, get infected.

A critical question for governments, however, is whether their populations do indeed adhere to social distancing practices once asked to do so. With adequate network analytics, mobile telecom operators can provide the answer to that – and a little more.

Mobile operators can, for example, evaluate the change in the mobility radius of their subscribers. That means isolating the distinct cell sites used by each client on different days, determining the Euclidean centroid amongst all of them and then calculating the radius of mobility that each subscriber displayed. A reduction in the radius of all cell sites covered by each subscriber indicates a drop in their area of movement and an increased effort to ‘stay put’.

Analysis that use cell (CGI) locations derived from GPS-leakage based azimuth inference create a more representative depiction of user positions. Nevertheless, that information still carries a variable degree of inherent inaccuracy depending on the CGI density of each region. Also, some cell changes of stationary users may still be present in the network analytics data of mobile operators, but they are likely to be present throughout any assessment. As a result, their relative change of locations would still be a meaningful approximation of social distancing adoption.

Another proxy for movement restrictions comes from the usage of ride-hailing services. Mobile operators can assess changes in ride-hailing events triggered by their subscriber base (i.e. the volume and frequency of car rides requested), revealing with accuracy if people are indeed reducing their demand for transportation.

Changes in ride-hailing transactions differ from fluctuations in the total data volume generated by subscribers while using transportation apps. Considering the broad range of services offered by some of those platforms, only the detection of specific transactions can reveal precisely if the volume of rides is falling. When subscribers move around less, that is an indication that their likelihood of meeting other people (or spreading the virus to other areas) will decrease.

Finally, the fraction of people who change their home bases after social distancing kicks in (for example, by leaving their cities to weather the pandemic at summer houses) is also observable. For that, mobile operators must define for each subscriber the centroids of cell sites used during nighttime hours (in order to better reflect their home locations) and then identify the subscribers whose nighttime centroids change significantly (for example, by more than 100 Km) from one period to another.

Movement metrics derived from network analytics data can offer a clear-cut estimate of social distancing adoption.

There is still little reference to judge, however, if a specific drop in digital personal mobility is enough to temper COVID-19’s growth. When epidemiologists gain a better understanding on the correlation between the movements detected by network analytics and effective viral spreading, the target thresholds for metrics produced by mobile operators will become even more evident.