CMStatistics 2021: Start Registration
View Submission - CMStatistics
B1622
Title: Stochastic time-discrete SIR models and particle filtering Authors:  Koichi Yano - Komazawa University (Japan) [presenting]
Tae Okada - Komazawa university (Japan)
Abstract: A novel method is proposed for estimating a stochastic time-discrete SIR (Susceptible, Infectious, or Recovered) model using particle filtering with time series of the number of infected and recovered persons. The time-discrete/time-continuous SIR model, which is one of the compartmental models of epidemiology, is a prevailing model in theoretical epidemiology. By combining the model with state and parameter estimation, the more precise forecasting of the number of infected persons is realized. My new method, which combines state estimation based on particle filtering and parameter estimation based on the Nelder-Mead method/the particle Metropolis-Hasting method, can reproduce the actual time series of the number of infected persons more faithfully. Using it, we estimate a stochastic time-discrete SIR model with the daily COVID-19 time series for Japan, simulate the number of infected persons in Japan, and forecast the number of infected people based on the estimated model. In addition, we analyze the factors behind the spread and decrease of COVID-19 infection in Japan based on the estimated states and parameters. This research is a great contribution to the global economy because the method will enable governments to quickly implement better infection control measures.