Analyzing the Impact of Covid-19 Control Policies on Campus Occupancy and Mobility via Passive WiFi Sensing

Camellia Zakaria, Amee Trivedi, Michael Chee, Prashant J. Shenoy, Rajesh Balan

Abstract

This paper investigates the utility of existing campus WiFi infrastructure for passive monitoring of indoor occupancy and human mobility during the COVID-19 pandemic. By analyzing network logs across three universities in the US and Singapore, the authors demonstrate that while partial restrictions (e.g., split-teams) reduce overall campus density, only strict lockdowns effectively curb individual mobility. This approach provides a privacy-preserving, infrastructure-based alternative to low-adoption contact tracing apps for institutional crowd control.

Introduction

When the COVID-19 pandemic emerged, universities implemented various safety policies, including remote learning, split-teams, and social distancing. However, monitoring compliance with these rapidly changing rules proved challenging. Traditional digital solutions often rely on "client-centric" mobile apps using Bluetooth or GPS. These methods face significant hurdles: GPS is inaccurate indoors, Bluetooth requires new hardware deployment, and both necessitate active user participation, leading to low adoption rates. This study addresses these gaps by exploring whether existing WiFi networks can passively monitor crowd density and movement without requiring user action.

Methodology: Network-Centric Passive WiFi Sensing

The researchers utilized a network-centric passive WiFi sensing approach, pulling data directly from enterprise WiFi infrastructure (Aruba and Cisco routers). When a smartphone has WiFi enabled, it pings nearby Access Points (APs) to associate or scan for networks. By collecting these system logs ("syslogs"), the researchers tracked two key metrics: occupancy and mobility.

Comparison of Client-Centric and Network-Centric sensing attributes.

To protect user privacy, the system does not access the user's device. All identifiable information is anonymized using a consistent SHA-2 hash before analysis, allowing for aggregate crowd flow tracking without identifying individuals. As shown in the comparison above, network-centric sensing requires no user input and leverages existing infrastructure.

Dataset and Campus Profiles

The study leverages real-world WiFi network logs from three university campuses, capturing transitions from pre-COVID operations through various lockdown phases starting in February 2020. The scale of these deployments is summarized below:

Summary of campus datasets used in the study.

The campuses include:

  • Singapore Management University (SMU): A non-residential city campus (~7 buildings, ~10,000 students/staff).
  • National University of Singapore (NUS): A large residential campus (~240 buildings, ~50,000 students/staff).
  • University of Massachusetts Amherst (UMASS): A large residential campus in the US (~230 buildings, ~38,000 students/staff).

Analysis of Policy Impact

The study distinguishes between local spread (virus transmission in dense rooms) and mobility spread (transmission via travel to new locations).

Key insights include:

  • Partial Policies and Mobility: Policies like hybrid classes or A/B split-teams effectively reduce overall campus occupancy (curbing local spread) but do little to reduce the mobility of individuals on campus. Students still visit multiple locations, maintaining a high risk of mobility spread.
  • The Lockdown Trade-off: Strict stay-at-home orders were the only policies that successfully reduced mobility spread. However, this shifted local spread risk to residential areas, where dorm occupancy skyrocketed, creating dense environments vulnerable to outbreaks.
  • Infrastructure as a Sensor: Coarse-grained data from AP locations is sufficient for macro-level policy evaluation and crowd control, proving that exact GPS coordinates are not necessary for institutional safety monitoring.

Results

  • Occupancy Reductions: SMU saw a >90% drop in occupancy during work-from-home phases, reaching nearly 100% during full lockdown. UMASS experienced a similar >90% decrease.
  • Mobility Trends: During the UMASS lockdown, the median number of unique places visited by dorm residents dropped from ~10 to fewer than 5.
  • Micro-Mobility: Conversely, NUS data showed that students confined to dorms increased short-distance transitions (e.g., between rooms and dining halls) from an average of 3 to 6 per day.

Conclusion and Future Work

While effective, passive WiFi sensing has limitations, including under-counting (users without smartphones or with WiFi disabled) and coarse-grained location data. Future work involves the open-sourcing of the Elastic WiFiTrace dashboard and deeper analysis of cultural factors influencing mobility patterns across different global regions.

Summary updated

BibTeX
@article{zakaria2022analyzing,
  author = {Camellia Zakaria and Amee Trivedi and Michael Chee and Prashant J. Shenoy and Rajesh Balan},
  title = {{Analyzing the Impact of Covid-19 Control Policies on Campus Occupancy and Mobility via Passive WiFi Sensing}},
  journal = {ACM Trans. Spatial Algorithms Syst.},
  year = {2022},
  doi = {10.1145/3516524},
}
Figure