Robust Inference in SEIR Models: A Lifebelt Particle Filter Approach

Dhorasso Junior Temfack Nguefack, Trinity College Dublin

Co-authors: Jason Wyse, Trinity College Dublin

Abstract: Emerging infectious disease outbreaks pose significant challenges for epidemiological modeling, particularly when dealing with sparse or noisy early-stage data. We propose an approach for state and parameter estimation in the Susceptible-Exposed-Infectious-Removed (SEIR) model by embedding the Lifebelt Particle Filter (LBPF) within the Sequential Monte Carlo Squared (SMC²) method. The LBPF addresses particle degeneracy by ensuring that at least one particle follows a wisely chosen trajectory, preventing particle collapse. SMC² performs joint state and parameter estimation as new data arrive. The proposed approach is used to estimate the time-dependent reproduction number during the 2018–2019 Ebola outbreak in the Democratic Republic of the Congo (DRC).