Abstract
The global spread of COVID-19 has profoundly affected health and economies, highlighting the need for precise epidemic trend predictions for effective interventions. In this study, we used infectious disease models to simulate and predict the trajectory of COVID-19. An SEIR (susceptible, exposed, infected, removed) model was established using Wuhan data to reflect the pandemic. We then trained a genetic algorithm-based SEIR (GA-SEIR) model using data from a specific U.S. region and focused on individual susceptibility and infection dynamics. By integrating socio-psychological factors, we achieved a significant enhancement to the GA-SEIR model, leading to the development of an optimized version. This refined GA-SEIR model significantly improved our ability to simulate the spread and control of the epidemic and to effectively track trends. Remarkably, it successfully predicted the resurgence of COVID-19 in mainland China in April 2023, demonstrating its robustness and reliability.
Introduction
Traditional mathematical models for epidemics, such as the standard SEIR model, often struggle in real-world applications because they treat populations as static variables. In reality, the spread of a virus is heavily influenced by the "human element"—behaviors like mask-wearing, social distancing, and the onset of "pandemic fatigue." Early in the COVID-19 pandemic, standard models frequently failed to account for rapid shifts in public policy and individual behavior, leading to inaccurate forecasts. This research addresses this gap by integrating socio-psychological indices directly into the model's differential equations, optimized via a Genetic Algorithm (GA).
Methodology: The GA-SEIR Framework
The researchers modernized the SEIR framework through a three-step approach:
- Standard SEIR Baseline: Establishing the basic mathematical flow of the disease using differential equations for Susceptible (S), Exposed (E), Infected (I), and Removed (R) populations.
- Genetic Algorithm (GA) Optimization: Instead of manual parameter estimation, a GA was employed to "evolve" the best-fitting parameters (e.g., infection and recovery rates) by minimizing the error between predicted and actual data.
- Socio-Psychological Integration: The core innovation involves two dynamic indices: Compliance (C), quantifying adherence to health guidelines, and Sentiment (S1), measuring public mood. These indices adjust transmission and removal rates in real-time.
Datasets and Validation
The study utilized three distinct datasets to refine and test the model:
- Wuhan City Data (2021–2022): Used for initial model construction and parameterization.
- Johns Hopkins University US Data: This dataset included quantified socio-psychological factors (compliance and sentiment), allowing the team to train and refine the GA-SEIR model.
- National Health Commission of China Data (April 2023): Used for external validation to test if the refined model could accurately predict future trends in a different geographic context.
Key Results and Insights
The GA-SEIR model demonstrated superior fit compared to standard models, particularly during critical peak periods. The Genetic Algorithm pinpointed optimal rates for the studied population: an infection rate of 3%, a conversion rate (exposed to infected) of 14%, and a removal (recovery/death) rate of 7%.
A major insight is that infection rates are social variables rather than biological constants. By treating "Compliance" as a scaling factor for transmission, the model becomes resilient to policy changes. Furthermore, the inclusion of public sentiment significantly improved the tracking of epidemic peaks, suggesting that public mood is a leading indicator of how a wave will resolve.
Conclusion
The refined GA-SEIR model provides crucial insights for public health authorities, enabling proactive strategies for outbreak containment and mitigation. While the model currently assumes a constant population and requires re-tuning for different pathogens, its ability to integrate human behavior marks a significant advancement in epidemic modeling. Future work will focus on increasing data granularity and accounting for human mobility to further enhance accuracy.
Summary updated