Congratulations to the winners of the FluSight Challenge — rodeo_flagellum, JonathanShi, and chipsie! Read the full announcement here.
We're proud of forecasters' efforts supporting public health officials working to mitigate the impact of flu season. Forecasts will be included in an upcoming manuscript by Professor Thomas McAndrew and colleagues that describes the methods used to scale human judgment forecasts.
2022-3-9 Update: Forecasters are now able to input/update multiple forecasts at once. Click each individual forecast question in the group, inputting your distribution for each one. Then, submit all your forecasts with a single button press.
Update: The FluSight Challenge now features question groups, allowing for more efficient, more consistent forecasting. Learn more about question groups here.
Welcome to the FluSight Challenge! This tournament asks the Metaculus Community to predict the intensity of the influenza season. Because the flu season’s peaks vary year to year, accurate forecasts can help public health officials navigate this uncertainty, plan more effectively, and reduce the flu’s impacts on their communities.
The Centers for Disease Control and Prevention (CDC) holds yearly FluSight challenges to solicit such forecasts from teams at institutions like Johns Hopkins, Columbia University, and Carnegie Mellon. This year, we’re excited to contribute predictions to the FluSight Challenge hosted by the CDC, while continuing our collaborations with Professor Thomas McAndrew’s Computational Uncertainty Lab at Lehigh University and Professor Evan Ray at the University of Massachusetts at Amherst.
How does this tournament work?
Participants will forecast weekly confirmed flu hospitalizations for six states: New York, Florida, Oklahoma, California, Vermont, and Wyoming. For each of these states, three questions will be open at all times and will solicit flu count predictions for dates spaced two weeks apart. These forecasts will provide inputs to a computational model developed by Professor McAndrew’s lab, which will then be used to project flu counts for every state and territory in the US. These projections will be submitted to the CDC’s FluSight Challenge and compared to predictions generated by the other competing teams.
For this tournament, forecasters will not be able to see the community median for the first 20% of a question's duration, and a forecaster’s coverage is determined by their participation during the hidden period. To learn more about the tournament scoring system, click here.
