Modelling the Spread of COVID-19 and the Effect of Preventive Measures : A Bayesian Hierarchical Latent Factor Approach
Horawala Withanage, Chathurangi (2025)
2025
Master's Programme in Computing Sciences and Electrical Engineering
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2025-07-31
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202507317937
https://urn.fi/URN:NBN:fi:tuni-202507317937
Tiivistelmä
The COVID-19 pandemic has posed a major global health challenge, requiring robust statistical tools to monitor and understand transmission dynamics. This thesis presents a Bayesian hierarchical latent factor model to study the evolution of the effective reproduction number (Rt) across four countries: Finland, Sri Lanka, Sweden, and India. The model captures both common global patterns and country-specific variations in transmission over time, offering a structured framework for cross-country comparison.
Two models are considered, one with a single latent factor and another with two factors, to compare how well they explain the observed patterns. The latent factors are modeled as time evolving processes, while country-specific loadings capture the sensitivity of each country to these shared trends. The framework incorporates serial interval distributions and detection probabilities, allowing for realistic transmission and reporting dynamics. Inference is performed using Markov chain Monte Carlo (MCMC) methods via the JAGS platform.
Results show that a single shared latent factor already captures much of the temporal variation across countries, reflecting the global nature of the pandemic. The two-factor model provides additional flexibility and identifies region-specific trends. However, the resulting correlation structures between countries were relatively uniform, suggesting that the dominant dynamics were largely synchronized across locations, or that data limitations reduced the model’s ability to distinguish local differences.
This work demonstrates how latent factor models, combined with hierarchical Bayesian methods, can be used to jointly analyze epidemic trends across multiple countries. The approach offers a flexible and interpretable way to assess both shared and country-specific dynamics in public health data.
Two models are considered, one with a single latent factor and another with two factors, to compare how well they explain the observed patterns. The latent factors are modeled as time evolving processes, while country-specific loadings capture the sensitivity of each country to these shared trends. The framework incorporates serial interval distributions and detection probabilities, allowing for realistic transmission and reporting dynamics. Inference is performed using Markov chain Monte Carlo (MCMC) methods via the JAGS platform.
Results show that a single shared latent factor already captures much of the temporal variation across countries, reflecting the global nature of the pandemic. The two-factor model provides additional flexibility and identifies region-specific trends. However, the resulting correlation structures between countries were relatively uniform, suggesting that the dominant dynamics were largely synchronized across locations, or that data limitations reduced the model’s ability to distinguish local differences.
This work demonstrates how latent factor models, combined with hierarchical Bayesian methods, can be used to jointly analyze epidemic trends across multiple countries. The approach offers a flexible and interpretable way to assess both shared and country-specific dynamics in public health data.