Forecasting Ebola with a regression transmission model

Abstract

We describe a relatively simple stochastic model of Ebola transmission that was used to produce forecasts with the lowest mean absolute error among Ebola Forecasting Challenge participants. The model enabled prediction of peak incidence, the timing of this peak, and final size of the outbreak. The underlying discrete-time compartmental model used a time-varying reproductive rate modeled as a multiplicative random walk driven by the number of infectious individuals. This structure generalizes traditional Susceptible-Infected-Recovered (SIR) disease modeling approaches and allows for the flexible consideration of outbreaks with complex trajectories of disease dynamics.

Keywords: Bayesian inference; Ebola; Forecasting; Mathematical modeling.