Introduction
Traditional contact tracing is often a race against time, hampered by human memory errors or low adoption rates of client-side apps. This research introduces a framework that leverages existing university Wi-Fi (WLAN) infrastructure to passively identify "superspreaders"—individuals whose social patterns make them critical nodes for disease transmission. Unlike previous methods, this approach accounts for environmental viral lingering through "asymmetric contact" and utilizes network centrality measures validated by SEIR simulations to prioritize public health interventions.
Methodology
The proposed framework transforms raw WLAN connection logs into actionable epidemiological data through a multi-stage pipeline.
Trajectory Reconstruction and Graph Construction
The process begins with Trajectory Reconstruction, where connection logs are sorted by timestamp to estimate stay durations at specific Access Points (APs). Walking times between buildings are estimated using the Google Maps API to create a continuous movement profile.
A key innovation is the development of Hybrid Contact Graphs, which move beyond simple simultaneous proximity.
- Symmetric Contact: Represents standard mutual proximity where two individuals are connected to the same AP at the same time.
- Asymmetric Contact: Accounts for "environmental infection." If an individual (e.g., Bob) stays at an AP for a significant duration and another (e.g., Trudy) visits shortly after Bob leaves, a directional edge is drawn from Bob to Trudy. This captures the risk of viral particles lingering in a shared space.
Identifying Superspreaders
To identify critical nodes, the researchers apply three network centrality measures:
- Degree Centrality: The total number of unique contacts.
- Closeness Centrality: How "near" an individual is to all others in the network.
- Betweenness Centrality: How often an individual acts as a "bridge" between different social clusters.
The effectiveness of these metrics is validated using a Susceptible-Exposed-Infectious-Recovered (SEIR) simulation. By "quarantining" top-ranked individuals according to each metric, the study measures which approach most effectively slows the epidemic's spread.
Dataset and Context
The framework was tested using a large-scale WLAN dataset from a major public university, comprising 139 days of logs from over 5,000 students and 8,000 Access Points. While the data (collected in 2015) predates the COVID-19 pandemic, it provides a high-fidelity map of "normal" campus interactions, essential for modeling viral transmission in dense populations.
Key Insights and Results
- The Importance of Asymmetry: Incorporating asymmetric contact tracing revealed "hidden" superspreaders missed by symmetric-only models, significantly improving the accuracy of the spread simulations.
- Betweenness as a Primary Metric: On a daily scale, Betweenness Centrality proved most effective. Individuals who move between different social groups (e.g., different departments) are more dangerous spreaders than those with many contacts within a single isolated group.
- Stability of Social Patterns: Student schedules stabilize quickly. By the third week of a semester, the identified superspreaders remain largely consistent, allowing for early-term identification and intervention.
- The 20% Intervention Rule: Simulations indicated that quarantining the top 20% of identified superspreaders provides the maximum reduction in total infections, with diminishing returns for higher quarantine rates.
Limitations and Future Work
The study notes that Wi-Fi logs track AP connections rather than precise 6-foot distancing, potentially leading to false-positive contacts in large rooms. Furthermore, the pre-pandemic dataset does not reflect behavioral changes like mask-wearing. Future research will focus on refining "overlap time" parameters and integrating this framework into real-time university dashboards for proactive pandemic management.
Summary updated