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

1 INTRODUCTION

• Institutions struggle to monitor compliance as public-health measures change, while existing indoor-sensing approaches can require additional deployment.The introduction identifies existing WiFi infrastructure as a feasible sensing resource for educational institutions.Source: §1 · p. 2
• The study analyzes anonymized, aggregated WiFi data from three universities, measuring occupancy with unique smartphone counts and mobility by the order of buildings users visit.Two campuses are in Singapore and one is in the northeastern United States.Source: §1 · p. 2
• Across three campuses in two countries, occupancy and mobility changed with restrictions; online learning reduced occupancy, while mobility stayed high among on-campus staff and students until remote learning.Mobility decreased significantly during the remote-learning period.Source: §1 · pp. 2–3
• The work evaluates policy effects and institutional compliance rather than solving a prediction problem or replacing digital contact tracing.The authors frame coarse-grained analysis as a privacy-preserving way to monitor aggregated congregation, occupancy, and crowd movement.Source: §1 · p. 3
• The paper provides evidence that passive WiFi monitoring of indoor occupancy and mobility could help administrators choose safe-distancing measures for similar outbreaks.Its policy analysis covers campus populations across two countries.Source: §1 · p. 3

2.1 COVID-19 College Outbreak

• By July 2020, 6,600 cases linked to 270 colleges had been reported before fall term, as universities faced movement restrictions, quarantine sites, and changing reopening guidance.Plans included decreasing occupancy in poorly ventilated areas, staggering use, and restricting occupancy in shared spaces.Source: §2.1 · p. 3

2 BACKGROUND AND MOTIVATION

• Location and tracking methods underpin digital COVID-19 responses, while adapting rules and monitoring institutional compliance create mobile-sensing opportunities.The background also reviews prior work on digital contact tracing and practical WiFi sensing for indoor monitoring.Source: §2 · p. 3, §2.2 · p. 4, §2.1 · p. 4

2.2 Digital Technologies for Monitoring COVID-19

• COVID-19 monitoring tools include trackers of infections and spread, alongside crowd-surveillance systems using video, footfall counters, and IoT sensors.The systems measure crowd density, movement, and facility usage.Source: §2.2 · p. 4

2.3 Leveraging WiFi-based Location Sensing

• WiFi is proposed as a complementary modality because low compliance with Bluetooth smartphone sensing motivates assessing its usefulness.Source: §2.3 · p. 4

2.3.1 Indoor Crowd Monitoring Systems.

• Existing COVID-19 WiFi methods for identifying at-risk occupants or superspreader events were demonstrated through simulated examples.Source: §2.3.1 · p. 5
• WiFi sensing can track indoor crowd change across areas and buildings, but cannot provide the finer-grained inter-user trajectories available from Bluetooth apps or cameras.The paper frames these aggregated occupancy and mobility patterns as useful for identifying potential exposure paths.Source: §2.3.1 · p. 5

2.3.2 Network-centric Sensing.

• Network-based sensing passively observes devices through WiFi infrastructure, whose AP logs can indicate nearby locations and campus occupancy.Unlike client-based GPS or Bluetooth sensing, it does not require an app or active user involvement.Source: §2.3.2 · pp. 5–6

2.3.3 Challenge #1: No WiFi Network.

• WiFi sensing assumes coverage wherever monitoring is needed; outdoor availability may vary with indoor AP placement.The stated monitoring focus is enclosed areas, where COVID-19 spread poses more risk to occupants.Source: §2.3.3 · p. 6

2.3.4 Challenge #2: Disconnected Users.

• Users without smartphones are overlooked, while cellular-data users may appear unconnected and visitor-device logs provide only coarse occupancy estimates.Unconnected-device records make precise monitoring challenging.Source: §2.3.4 · p. 6

2.3.5 Challenge #3: Coarse-grained Location.

• AP-level positions are adequate for aggregated occupancy trends and overall campus mobility, but not for precise contact tracing.Source: §2.3.5 · p. 7

2.4 Privacy and Ethical Considerations

• The proposed compliance monitoring uses anonymized network data at an aggregated scale rather than identifying individual occupants.If information is needed for contact tracing, only an authorized official may obtain a de-anonymized copy after an individual is identified as at-risk.Source: §2.4 · p. 7

• Every MAC address was hashed with a consistent one-way function, and the analysis focused on large aggregates rather than specific individuals.Source: Data Ethics · pp. 7–8

3.1 Data Collection

• The datasets come from three production campus WiFi networks: two large residential universities and one smaller non-residential university.The large campuses each have over 200 buildings and approximately 40K–50K students and staff; the smaller campus has about 10,000 across seven buildings.Source: §3.1 · p. 8
• The study collected AP identifiers, timestamps, and hashed client MAC addresses from associated devices, and also obtained such records for unassociated devices at two campuses.Data were pulled from Aruba and Cisco infrastructure through APIs or system logs.Source: §3.1 · p. 8
• Associated-device data for all three campuses begin in February 2020, covering periods before, during, and after COVID-19 measures.Table 2 provides details on the campuses and collected data.Source: §3.1 · p. 8, Table 2 · p. 8

3.2.1 Key Assumptions.

• The analysis assumes most users use campus WiFi on their smartphones; the cited UMASS analysis found about 90% of university residents carried a smartphone, while 10% neither owned one nor chose campus WiFi.The approach therefore measures some occupancy and mobility among smartphone users, not every campus user.Source: §3.2.1 · p. 8

3.2.2 Pre-processing.

• Timestamped AP syslog events are assembled into per-device connection sessions and trajectories, which support user profiles and activity estimates.The logs include association, disassociation, authentication, deauthentication, re-association, and drift events; MAC addresses are anonymized.Source: §3.2.2 · p. 9

3.2.3 On-campus Deployment.

• As a user moves, a smartphone switches among nearby APs, producing network events that trace movement across campus locations.Enterprise APs are distributed across buildings and floors; the main analysis uses associated devices representing staff and students.Source: §3.2.3 · p. 9
• A device may remain connected to an AP after its user moves, creating small mobility errors but not affecting floor- or building-level occupancy counts.Multiple devices per user can also duplicate events, which can be removed or filtered by device type.Source: §3.2.3 · p. 10

3.3 Inferring Occupancy and Mobility Using WiFi Logs

• WiFi-based occupancy uses AP-defined areas that may combine small rooms or correspond to one large room, so room-level location can be inaccurate.Floor occupancy sums the areas on a floor, and building occupancy sums the floors; users with one-time or infrequent logins are filtered as likely visitors.Source: §3.3 · p. 10

3.3.1 Definition: Occupancy.

• Occupancy is the average number of people in an area on a building floor, estimated from unique logged smartphones and separated into associated university residents and unassociated visitors.At Study Area 1, logs recorded 2 devices from 3:00–4:00 AM and 73 device connections from 1:00–2:00 PM on a Monday.Source: §3.3.1 · pp. 10–11, Figure 3 · p. 11

3.3.2 Definition: Mobility.

• Mobility is the average number of areas visited by a user per hour, inferred from WiFi trajectories as devices switch among APs.The analysis treats AP-to-AP movement as evidence of transitions; temporal lag between nearby APs does not affect building-level transition estimates.Source: Figure 4 · p. 11, §3.3.2 · p. 12
• SMU WiFi logs show building-to-building transitions vary by time: Building 7 drew 259 transitions from Building 2 during lunch, while later transitions were more distributed.The chord diagram covers all seven SMU buildings and compares movement at different times of day.Source: §3.3.2 · p. 12, Figure 5 · p. 12
• The resulting occupancy and mobility statistics can identify high-foot-traffic campus areas and support analysis of changes across COVID-19 safety measures.Table 3 summarizes the five phases of measures enacted in Singapore and at SMU and NUS.Source: §3.3.2 · p. 12, Table 3 · p. 13

4.1 Overall Control Policy

• Singapore’s measures escalated from visitor quarantine and online academic programs to country-wide stay-at-home orders, with dormitory blocks repurposed as quarantine centers.The analysis compares campus occupancy and mobility across the restriction phases.Source: §4.1 · p. 13, §4 · p. 13

4.2 Impact of Different Policies on Occupancy

• SMU occupancy dropped more than 90% from PS0 to PS3 and beyond, while NUS occupancy was reduced to 75% at PS3 and 98% by lockdown; SMU seminar-room use fell more than 50% on average after large classes moved online.SMU occupancy was almost zero by PS4, while NUS still had students living in dorms.Source: Figure 6 · p. 14, §4.2 · p. 14
• Space use differed by campus: SMU closed recreational facilities and mostly closed dining facilities from PS2, whereas NUS retained relatively high recreational occupancy during PS4.The authors suggest dorm residents’ continued use of recreation spaces may have led to mixing among students from different dorms.Source: §4.2 · pp. 14–15, Figure 7 · p. 15, Figure 8 · p. 15

4.3 Impact of Various Policies on Mobility

• At SMU and NUS, overall transitions decreased with campus occupancy, but mobility per person at NUS remained high until stricter lockdown measures began.SMU’s A/B schedules reduced occupancy and transitions; at NUS, staff and students remaining on campus continued visiting the same set of buildings. The figures compare transitions from one origin building to other buildings.Source: §4.3 · pp. 15–16, Figure 9 · p. 16, Figure 10 · p. 16

4.4 On-campus Living

• NUS dorm activity stayed at about 700 across phases, while the sampled residents’ average transitions doubled from about three in PS2 to about six in PS3 and PS4.During lockdown, WiFi use shifted toward nighttime; residents’ trips were often shorter movements among neighboring dorm areas and nearby dining, recreation, or bus-stop locations.Source: Figure 11 · p. 17, Figure 12 · p. 17, §4.4 · pp. 17–18

4.5 Main Takeaways

• The authors report that telecommuting and split-team policies reduce campus density, but people who remain are likely to keep visiting familiar places; limiting movement may require providing essentials on-site.The dorm analysis notes that movement can rise even when trip durations are shorter if residents travel elsewhere for food and other necessities.Source: §4.5 · p. 18

5 MOBILITY AND OCCUPANCY ANALYSIS FOR US UNIVERSITY

• UMASS moved directly from business as usual to a stay-at-home order with business shutdowns, rather than through multiple policy phases.The occupancy comparison covers ten days in each phase.Source: §5 · p. 18
• Occupancy fell by more than 90% from business-as-usual phase P_U0 to full stay-at-home phase P_U1, with dormitories becoming the most occupied locations.Figure 13 breaks occupancy down by area type.Source: §5 · p. 18, Figure 13 · p. 19
• At the 50th percentile, visits to unique locations fell from about ten in P_U0 to fewer than five in P_U1; the 90th percentile fell from about 17 to about nine.The CDF compares occupants of UMASS_Dorm before COVID-19 with the quarantine phase.Source: §5 · p. 19, Figure 14 · p. 19
• Most P_U0 visits were to dining and recreation, while most P_U1 visits were to another on-campus dorm, which the authors attribute to closures of previously popular campus locations.The percentage-change figure breaks visits down by place type.Source: §5 · p. 19, Figure 13 · p. 19

6 POLICY IMPACT

• The analysis distinguishes local spread among people in the same place from mobility spread involving travel to other places; the two vectors require different control approaches.Reducing density can also reduce mobility spread, though the authors note that limiting movement within a primary location may not always be practical.Source: §6 · p. 20

6.1 Controlling Local Spread Across Campus

• Quarantine removed people from workplaces, but student dormitory density initially rose; asking students to leave reduced campus density while shifting crowding toward homes and other residences.The authors note the potential for higher local infection risk in crowded residential settings.Source: §6.1 · p. 20

6.2 Controlling Mobility Spread

• Only strict quarantine significantly reduced individual mobility; split-team and other earlier approaches did not, and mobile people spent less time at each place visited.Mobility is defined as the number of unique places an individual visits in a day.Source: §6.2 · p. 21
• The authors argue that strict quarantine reduces the chance of spread beyond a local area, but may increase density and local infection risk in homes, dormitories, and other residential areas.They cite the Singapore worker dormitory outbreak as an example.Source: §6.2 · p. 21

7 DISCUSSION

• The authors frame coarse-grained WiFi data as a means to reflect crowd changes under different COVID-19 policies.They describe occupancy and mobility monitoring as a potential aid to campus operations.Source: §7 · p. 21

7.1 Support Operational Strategies for Universities

• WiFi-based reports of space use, occupancy, and foot traffic can help institutions identify crowded areas, monitor capacity, and plan gathering controls, emergency evacuation, and other campus operations.The proposed monitoring spans buildings, floors, and areas, with user types also potentially informing capacity decisions.Source: §7.1 · p. 21, Figure 16 · p. 22

7.2 Practical Application for Institutional Crowd Controls

• The dashboard uses unique WiFi MAC addresses to calculate occupancy percentages and displays area-level occupancy by time of day; alerts and capacity adjustments can inform crowd management and event scheduling.The authors report an open-source platform for setting occupancy parameters and mobilizing contact tracing.Source: §7.2 · p. 22

7.3 Implementation of Privacy Safeguards

• The system measures occupancy with anonymized WiFi data, while the authors recommend notice and consent before administrators access identifying information for manual contact tracing; campus managers at SMU and NUS receive occupancy and mobility updates.The user dashboard includes a pane listing anonymized occupants.Source: §7.3 · pp. 22–23, Figure 17 · p. 23

8 CONCLUSION & FUTURE WORK

• Across two Singapore campuses and one in the Northeastern United States, WiFi network information revealed occupancy and mobility changes under containment policies.The authors say augmenting WiFi data can assist institutions with safety compliance and provide an open-source monitoring tool.Source: §8 · p. 23
• The study may undercount people without WiFi-enabled devices associated with campus networks, and WiFi sensing cannot resolve small-room occupancy or inter-user distances.The authors say it instead identifies occupants’ degree of risk from being in an exposed area.Source: §8 · p. 23
• The findings come from two countries, and local factors may limit how specifically the results apply to other regions.The authors describe deeper and broader analyses as future work to address these shortfalls.Source: §8 · p. 23