Abstract
SEIR-Campus is a high-performance Python package designed to simulate the spread of infectious diseases within the unique, highly structured environment of a university. By combining agent-based modeling with real-world data like course registrations, it allows administrators and researchers to stress-test intervention policies—such as hybrid scheduling, mass testing, and contact tracing—in a matter of seconds. This paper details the model's nine-state framework and its computational efficiency in handling large-scale campus populations.
Introduction
Modeling an epidemic on a college campus is a "small-world" problem with a twist. Unlike general population models, student interactions are dictated by rigid schedules (classes), high-density living (dorms), and fluid social circles. Traditional epidemiological models often rely on static networks or differential equations that struggle to capture these day-to-day dynamic shifts. Existing tools are either too generalized or computationally sluggish when scaled to tens of thousands of individuals. The authors developed SEIR-Campus to fill this gap, providing a tool that is both granular enough to model individual student movements and fast enough to run hundreds of "what-if" scenarios quickly.
Methods
The package utilizes an agent-based SEIR (Susceptible-Exposed-Infectious-Removed) model with several sophisticated extensions designed for the university context.
Nine-State Framework
Beyond the standard four states, the model tracks symptomatic vs. asymptomatic infectiousness and mirrors these states for individuals in quarantine. This allows for realistic modeling of "self-reporting" behavior and the tracking of "false positives" (susceptible individuals in quarantine) versus "true positives" (infectious individuals in quarantine), which is vital for calculating the academic cost of a policy.
Meeting-Based Transmission
Instead of calculating infections by checking every possible pair of students (edge-based), the model processes "meetings" (classes or social events). It uses Poisson and Binomial processes to determine transmissions within a group, which is mathematically equivalent but computationally far superior for dense networks. The model uses "effective minutes" to calibrate risk; for example, a long lab session with social distancing can be modeled as having a shorter "effective duration" than a shorter, more crowded social gathering.
Data and Stochasticity
The model is demonstrated using the Weeden-Cornwell Network, based on anonymized transcript data from Cornell University. Key variables like the incubation period and recovery time are treated as stochastic processes, capturing the natural variability of biological viruses.
Results
SEIR-Campus exhibits high performance, simulating a population of 20,000 individuals for an entire semester in less than five seconds. This speed is critical for performing sensitivity analyses.
Key policy insights include:
- Intervention Efficacy: Moving classes with more than 50 students online or implementing a "hybrid" alternating-week schedule significantly flattens infection peaks.
- Testing Sensitivity: Simulations demonstrate that even with 75% of cases being asymptomatic, regular surveillance testing (e.g., testing 20% of the student body daily) can effectively contain an outbreak that would otherwise spiral out of control.
Discussion and Limitations
While SEIR-Campus provides a robust framework for campus modeling, it currently assumes equal exposure within a single "meeting" and supports only a single disease with lasting immunity. Future work may include modeling competing diseases (like the flu), waning immunity, and more granular spatial tracking. However, the modular design allows for "pluggable" interventions, making it a flexible tool for university public health planning.
Summary updated