Dynamic Bayesian Influenza Forecasting
Forecasts of seasonal influenza assist in public health planning and intervention. Challenges in producing these forecasts include limited and noisy data, and incomplete models of disease transmission leading to discrepancy of models from reality. This talk will discuss a dynamic Bayesian influenza forecasting approach, using the constraints of the domain drawn from subject matter models and historical observations. Data from a current flu season is used to forecast, with uncertainty, the likely values of quantities of interest of the future flu progression. This statistical modeling approach has led to a successful entry in the US Center for Disease Control (CDC) flu forecasting challenge, and, being complementary to other approaches, also provides improvement as a member of combined models. This talk will describe the approach to creating a dynamic online statistical forecasting system that incorporates domain models, observed data, and model discrepancy, it’s evaluation and comparison to other models in operation, and some current and potential future extensions.
Dr. James Gattiker Los Alamos National Laboratory