Real-time forecasting of an epidemic outbreak: Ebola 2014/2015 case study

Forecasting an epidemic outbreak is important for planning and interventions of health authorities. This paper presents a study of particle filtering based algorithms for estimation and forecasting the number of Ebola cases, using the World Health Organisation (WHO) observations collected during the 2014/2015 epidemic in West Africa. The dynamics of the Ebola epidemic is modelled using a stochastic and discrete version of the susceptible, exposed, infected, removed (SEIR) compartmental model. The likelihood function of the observed number of cases is adopted as a negative binomial distribution with the variance as a free parameter. Progressive correction is used to make the method more robust against data irregularities. For the sake of comparison and performance assessment, an error performance measure is defined as the Bhattacharyya distance between the reported number of cases and the estimated/forecasted number of cases.