Friday, May 6, 2016 - 8:00am to 9:00am
Location:Traffic 21 Classroom 6501 Gates & Hillman Centers
Speaker:LOGAN BROOKS, Ph.D. Student /LOGAN%20BROOKS
Seasonal epidemics of influenza cause significant morbidity, mortality, and economic loss; Centers for Disease Control and Prevention (CDC) estimates that they are associated with over 200,000 hospitalizations and 3,000 to 49,000 deaths in the United States each year. Accurate and reliable forecasts of disease prevalence can assist the design of more effective countermeasures and improve hospital and public preparedness. Unfortunately, while the number of influenza infections over time within a given season follows a rough pattern, with one or two sharp peaks occurring between December and March, it is not well modeled by simple time series tools such as linear autoregression, and calls for a more tailored approach. I present one of the CMU Delphi group's systems for forecasting disease prevalence, an ensemble of empirical Bayes, regression, and kernel-based methods, and analyze its cross-validated performance when forecasting doctor visit data for influenza-like illness (ILI).
Joint work with David Farrow, Sangwon Hyun, Shannon Gallagher, Roni Rosenfeld, and Ryan Tibshirani
Presented in Partial Fulfillment of the CSD Speaking Skills Requirement.