Introduction
Epidemiological modeling is a cornerstone of public health planning, yet traditional approaches often struggle with a trade-off between simplicity and realism. Simple compartmental models (e.g., SEIR) lack spatial and behavioral granularity, while many Agent-Based Models (ABMs) rely on oversimplified "synthetic" behaviors. AVSim addresses these gaps by introducing a novel agent-based simulation framework grounded in empirical GPS data. By capturing high-resolution human mobility patterns across diverse occupations, AVSim enables precise modeling of both airborne (e.g., COVID-19) and vector-borne (e.g., Dengue) disease dynamics.
Methodology
The AVSim pipeline consists of several integrated stages designed to translate raw mobility data into realistic agent behaviors.
- Mobility Pattern Clustering: Raw GPS data is processed using DBSCAN (Density-Based Spatial Clustering of Applications with Noise) to identify "stay regions"—locations where individuals spend significant time—while filtering out transient travel noise.
- Behavioral Subtyping: To account for heterogeneity within occupational groups (e.g., different work shifts), the framework employs Spectral Clustering. This identifies latent behavioral patterns, distinguishing between weekday and weekend routines or varied professional schedules.
- Environment Modeling: The simulation uses a hierarchical tree structure to represent geography, ranging from broad districts down to specific functional zones like schools, hospitals, and residential areas.
Probabilistic Trajectory Generation
A core innovation of AVSim is its stochastic yet statistically grounded approach to agent movement. Instead of hard-coded paths, the framework utilizes two distinct probability matrices: a Location Visit matrix ($L$) and a Location Occupancy matrix ($S$).
As shown in Figure 6, agents navigate the environment by selecting their next destination from matrix $L$ and their stay duration from matrix $S$ using "roulette wheel selection." This dual-matrix approach captures the "bursty" nature of human movement—where people go and exactly how long they stay—more accurately than standard random walk or simple gravity models.
Hybrid Disease Modeling
AVSim features a hybrid modeling approach to accommodate different transmission vectors:
- Airborne Transmission: Modeled through direct contact within a 1-meter radius between agents.
- Vector-Borne Transmission: Employs a "patch-based" approach. To maintain computational efficiency, the environment is divided into patches where mosquito density and infection rates are tracked via differential equations, rather than simulating individual vectors. This model also accounts for climatic factors like temperature and rainfall, which influence mosquito life cycles.
Results and Discussion
The framework was validated using a dataset of GPS traces from 100 volunteers in Kandy, Sri Lanka, covering 13 professions over 21 days. Key findings include:
- Occupational Hubs: Specific workplaces act as critical bridges for transmission between disparate communities (e.g., a student bringing a virus home to a parent who works in a hospital).
- Intervention Efficacy: Simulations showed that profession-based quarantine combined with self-quarantine could reduce the COVID-19 infection peak by 87.5%.
- Vector Control: For Dengue, the model demonstrated that even 75% effective vector control (e.g., mosquito spraying) can significantly flatten the outbreak curve if triggered early by a threshold of reported cases.
Conclusion
AVSim provides a robust, data-driven framework for simulating complex disease dynamics. By integrating empirical mobility data with behavioral subtyping and hybrid transmission models, it offers a powerful tool for testing targeted public health interventions. Future work will focus on scaling the agent population and incorporating more complex dynamics like waning immunity and non-compliant quarantine behaviors.
Summary updated