Abstract
Effective epidemic management requires a delicate balance between public safety and economic activity. Traditional risk assessment methods often rely on coarse city-level travel records or error-prone self-reporting. This paper proposes a novel two-stage framework that utilizes granular mobile communication station (cell tower) data to provide high-precision risk profiles. By classifying individual "passing modes"—such as driving, walking, or living—the system distinguishes between high-risk contact and low-risk transit. Simulations involving 10,000 users demonstrate that this method can achieve 100% accuracy in identifying high-risk companions while significantly reducing false positives for transiting individuals.
Introduction
During the COVID-19 pandemic, a critical challenge emerged: how to accurately assess individual exposure risk without resorting to broad, city-wide lockdowns that paralyze the economy. Existing tools generally lack the necessary granularity. For instance, a person driving through a high-risk zone in a sealed vehicle has a fundamentally different risk profile than someone walking through a crowded market in the same area.
This research addresses this gap by leveraging existing mobile infrastructure to map trajectories with high spatial and temporal precision. By moving beyond simple location-based tracking to behavior-based assessment, the proposed method offers a "fine-grained" approach to epidemic monitoring.
Methodology
The authors introduce a two-stage framework designed to process telecommunications data into actionable risk metrics.
Stage 1: Communication Station Risk Assessment
Rather than labeling entire districts as high-risk, the system calculates risk values for individual communication stations at 10-minute intervals. A key component is Passing Mode Classification, which weights the risk based on how an individual interacts with a station's range. Five modes are defined: Drive, Walk, Stop, Live, and Work. Mathematical weights ((\mu)) are assigned to these modes; for example, "living" or "working" near an infected individual carries a higher risk weight than "driving" past them.
Stage 2: Personnel Risk Assessment
This stage cross-references individual trajectories with the calculated station risks. It employs a Companion Judgement Method to detect if a user moved in tandem with a known infected person. A companion is identified if they share the same passing mode at the same station within a specific time window.
Key Insights and Trajectory Analysis
The core innovation of this method lies in its temporal and spatial granularity. By dividing time into 600-second (10-minute) slots, the system captures fleeting contacts that traditional data would overlook. Furthermore, the spatial analysis of trajectories through overlapping cell tower ranges allows for precise mapping of user movement.
As shown in the trajectory visualization, users ((U_i)) move through various station sectors ((S_j)). The system doesn't just look for overlapping locations; it looks for overlapping behaviors. Two people "walking" through the same sector simultaneously are flagged as higher risk than one person "walking" while another is "driving" through.
Experimental Results
Due to privacy concerns regarding real-world telecommunications data, the framework was validated using extensive simulations:
- Scale: 10,000 users were modeled, including 2,000 infected individuals, across 13 distinct sub-areas (urban, industrial, and suburban).
- Accuracy: The system correctly identified 100% of the 800 simulated "companions" as high-risk (scoring between 0.5 and 1.0).
- Specificity: For 4,000 users with non-overlapping trajectories, the system correctly assigned near-zero risk. For those who overlapped with infected areas but were in low-risk modes (like driving), the system provided a gradient of risk that avoided unnecessary flagging.
Conclusion and Future Work
The proposed communication station-based method provides a scalable, high-precision alternative to traditional epidemic tracking. By accounting for the nature of contact through passing modes, it offers a more nuanced view of public health risks. Future research will focus on transitioning from rule-based classification to Deep Neural Networks (DNNs) to better learn complex movement patterns and further improve classification accuracy in noisy, real-world signal environments.
Summary updated