Stochastic Epidemic Modelling

TL;DR

This paper explores stochastic epidemic modeling using hidden Markov Chains and particle MCMC algorithms to estimate model parameters and assess algorithm sensitivity under limited and imperfect data conditions.

Contribution

It introduces a novel approach combining hidden Markov Chains and particle MCMC with adaptive algorithms to improve epidemic parameter estimation from scarce data.

Findings

Effective modeling of epidemic scenarios with imperfect data.

Adaptive MCMC algorithms enhance parameter estimation accuracy.

Sensitivity analysis reveals robustness of the methods under data limitations.

Abstract

Inferring how an epidemic will progress and what actions to take when presented with limited information is of critical importance for epidemiologists and health professionals. In real world settings, epidemiology data can be scarce or subject to reporting errors. In this project there are different epidemic scenarios simulated and, using hidden Markov Chains, it is attempted to mimic the imperfect data an epidemiologist will encounter. Furthermore, different kinds of compartmental models are modelled using the particle Markov Chain Monte Carlo algorithm with a variation of the adaptive Metropolis-Hastings algorithm to estimate the posterior density of the parameters underlying the models. Moreover, the sensitivity of these algorithms is investigated when subjected with changes in the dataset. This is accomplished by limiting the information provided, while using an adaptive approach on the posterior covariance of the parameters.