Keywords: Bayesian Generalized Linear Model, Data Augmentation, Dynamically Weighted Importance Sampling, Radial Basis Function Networks, Spatio-temporal model
Global spread of the Crimean-Congo Hemorrhagic Fever (CCHF) is a fatal viral infection disease found in parts of Africa, Asia, Eastern Europe, and Middle East, with a fatality rate of up to 30%. A timely prediction of the prevalence of CCHF incidents is highly desirable while CCHF incidents often exhibit nonlinearity in both temporal and spatial features. However, the modeling of discrete incidents is not trivial. Moreover, the CCHF incidents are monthly observed in a long period and take a nonlinear pattern over a region at each time point. Hence, the estimation and the data assimilation for incidents require extensive computations. In this paper, using the data augmentation with latent variables, we propose to utilize a dynamically weighted particle filter to take advantage of its population controlling feature in data assimilation. We apply our approach in an analysis of monthly CCHF incidents data collected in Turkey between 2004 and 2012. The results indicate that CCHF incidents are higher at Northern Central Turkey during summer and some beforehand interventions to stop the propagation are recommendable.