Click here for the code for the paper "A non-parametric Hawkes model of the spread of Ebola in West Africa".

Click here for point process model Central Africa Ebola forecasting files .

In order to refine and improve models for forecasting the spread of contagious diseases like Ebola, it is necessary to engage in the most honest and rigorous evaluation of these models. Retrospective analyses tend to suffer from the problems of overfitting, publication bias, and related problems such as forking paths, vaguely specified tuning parameters, etc. Prospective studies suffer from these problems far more rarely.

Here, we put code and forecasts online and timestamped them to verify that no changes were made retroactively. The data are aggregates of the World Health Organization (WHO) outbreak reports on Ebola. Following Althaus (2014), the focus here was on counts of infection cases from Ebola in Southeast Guinea, Eastern Sierra Leone, and Northwest Liberia. The models fit here are the nonparameteric Hawkes model of Chaffee et al. (2019) and the recursive model of Schoenberg et al. (2018). We welcome others.

The format of the files is explained here by example. In the folder 2019-05-06_ebola, the data go up to May 6, 2019. Also in the same folder are the fitted model parameters and forecasts for the next 7, 14, and 21 days. The data used for fitting are contained in 2019-05-06_data.csv. The parameters of fitting the Hawkes and Recursive models forecasts are, respectively, in the files hawkes_params.txt and recursive_params.txt. For the Hawkes model forecasts, 1000 simulations of every model were run and the file hawkes_projection.txt contains the mean and sd of the 1000 outcomes. For the Recursive model forecasts, 50 simulations of every model were run and the file recursive_projection.txt contains the mean and sd of the 50 outcomes.

In the folder Code, the R code used is in the files (1) daily_data_format.R, (2) MarsanR-temporal.R, (3a) recursive.R, and (3b) fitdailyebola.R.

Althaus, CL (2014). Estimating the reproduction number of Ebola virus (EBOV) during the 2014 outbreak in West Africa. in PLOS Current Outbreaks, 2014; 6.

Chaffee A., Park J., Harrigan R., Krebs A., and Schoenberg F.P. (2019). A non-parametric Hawkes model of the spread of Ebola in West Africa. In review.

Schoenberg, F.P., Hoffmann, M., and Harrigan, R. (2019). A recursive point process model for infectious diseases. AISM , to appear.