Location
MC 6460
Speaker
Professor Monica Cojocaru, the College of Computational, Mathematical and Physical Sciences, University of Guelph
Title
Expanding optimization ensemble model methods for forecasting seasonal influenza in the U.S.
Abstract
Each year, the seasonal influenza epidemic sees large variability in its evolution. Having accurate forecasts of future influenza cases is important for planning public health response. The United States Centers for Disease Control and Prevention (CDC) has annually organized the FluSight competition (https://github.com/cdcepi/FluSight-forecast-hub) to solicit forecasts from participating teams. Using this data, the CDC produces an ensemble forecast by taking the median of all submitted forecasts, which ranks among the top forecasts in terms of performance. We introduce two new weight-based ensemble forecasting methods to consider predicting laboratory-confirmed influenza hospital admissions for the 2023-2024 season. The first method, expanding window optimization (EWO), consists of determining optimal weights to minimize the mean square error of a blend of teams' previous forecasts compared to the truth data, and using these weights to produce future forecasts. We then modify this method to produce adjusted weighted-EWO (adw-EWO) by incorporating an additive correction based on past performance. We observe that EWO is unable to improve on the FluSight Ensemble forecasts, whereas adw-EWO offers an improvement for the 1-week and 2-week ahead forecasts for certain fixed values of our correction parameter $\pi$. Furthermore, adapting the value of $\pi$ throughout a season shows potential to further improve forecasting accuracy and outperform the FluSight Ensemble method at all horizons.
This is joint work with: Benjamin Benteke, Rhiannon Loster, Pengfei Yue, David Lyver, Christopher M. van Bommel, Edward W. Thommes