Data-driven Infectious Disease Ecosystem Modeling

<p dir="ltr">The COVID-19 pandemic underscored the importance of developing holistic frameworks that incorporate a variety of external factors into infectious disease modeling, in order to guide public health responses. While traditional models often focus on isolated aspects of disease dynamics, this thesis proposes a unified infectious disease ecosystem framework that captures the complex interplay of non-pharmaceutical interventions (NPIs), pharmaceutical interventions (PIs), human mobility patterns, weather conditions, and pathogen variants. By bridging compartmental models such as SEIR (susceptible–exposed–infected–recovered) with machine learning techniques, this research addresses critical gaps in the literature and advances both theoretical understanding and practical applications in pandemic modeling. The thesis is structured around four interconnected research objectives, each contributing to the unified framework. First, it quantifies the waning effectiveness of NPIs over time by introducing a modified SEIR model, SVEIC-NLC, which embeds Newton’s Law of Cooling (NLC) to represent NPI fatigue; the model is validated across multiple countries. Second, it explores the behavioral heterogeneity induced by vaccination certificate (green pass) policies through the SEIQRD^2 model, demonstrating how these policies influence group dynamics and transmission rates. Third, it examines the critical role of human mobility in disease transmission by integrating mobility data with NPIs and PIs in a hybrid SEIR-neural network framework, NeuralSEIR, to reveal its influence on pandemic progression. Finally, the thesis develops SEIR-VRNN, a comprehensive model that simultaneously incorporates NPIs, PIs, human mobility, and weather conditions using a variational recurrent neural network (VRNN), enabling robust forecasting and scenario analysis. By synthesizing these contributions, the thesis advances a unified knowledge framework for infectious disease modeling that integrates dynamic external factors, balances predictive accuracy with interpretability, and provides actionable insights for public health planning. Validated using real-world data from diverse regions, the proposed models demonstrate scalability, adaptability, and relevance for managing current and future pandemics. This work lays the foundation for a new generation of integrative models capable of addressing the multifaceted challenges of infectious disease dynamics.</p>