St. James Ballroom
Modeling Malaria Prevalence: Comparison of Multiple Spatial and Spatio-Temporal Models (303910)
*Ben Toh, University of Florida
Denis Valle, University of Florida
Keywords: malaria,spatial,spatio-temporal,prediction
Despite advances in control and prevention, malaria is still one of the major public health challenges in Africa. Malaria survey data is increasingly abundant in bid to eliminate the disease. Geospatial statistical models play important role: they are key to produce reliable malaria risk maps. Many models have been used in the literatures, and it poses challenges to the modelers: what is the best model? Where multiple years of survey data are available, modelers also have to choose whether to include past data in the model. Model comparison studies are essential to guide the modelers, but such studies are limited. In this study, we compare the predictive performance of various models, with or without incorporating past data, in predicting malaria prevalence in five African countries. We use up to 10 predictors derived from geographical and remote sensing products, and 5 models: GLM, GAM, BRT, BART and SPDE-INLA. Generally, having past data improves models' performance. However, there were no clear consensus in what is the best model. We discuss the pros and cons of the methods, e.g. computation time, difficulty in learning and using software packages and etc.