arXiv:2406.19524v1 Announce Kind: new
Summary: Agent-based fashions (ABM) present a superb framework for modeling outbreaks and interventions in epidemiology by explicitly accounting for various particular person interactions and environments. Nevertheless, these fashions are often stochastic and extremely parametrized, requiring exact calibration for predictive efficiency. When contemplating real looking numbers of brokers and correctly accounting for stochasticity, this excessive dimensional calibration might be computationally prohibitive. This paper presents a random forest primarily based surrogate modeling method to speed up the analysis of ABMs and demonstrates its use to calibrate an epidemiological ABM named CityCOVID by way of Markov chain Monte Carlo (MCMC). The method is first outlined within the context of CityCOVID’s portions of curiosity, specifically hospitalizations and deaths, by exploring dimensionality discount by way of temporal decomposition with principal part evaluation (PCA) and by way of sensitivity evaluation. The calibration downside is then offered and samples are generated to greatest match COVID-19 hospitalization and loss of life numbers in Chicago from March to June in 2020. These outcomes are in contrast with earlier approximate Bayesian calibration (IMABC) outcomes and their predictive efficiency is analyzed exhibiting improved efficiency with a discount in computation.
Supply hyperlink