Skip to content
Abhinav Tushar edited this page Jan 29, 2018 · 9 revisions

DOI Badge

The instance of flusight deployed at http://reichlab.io/flusight displays live flu predictions for U.S.A. from Reich Lab at UMass Amherst. The predictions are based on live data provided per week by Centers for Disease Control and Prevention in form of weighted ILI (wili%). Other efforts for predicting the season's flu trend can be seen at Epidemic Prediction Initiative's website, the DELPHI group's forecast visualization site, and the Columbia Prediction of Infectious Disease site.

Recommended Citation:

Tushar A, Reich NG, Ray EL. flusight: Influenza Forecasts Visualizer (v2.0). 2017. DOI 10.5281/zenodo.888171.

Visualization

Flusight displays interactive visualization of prediction models along with the confidence intervals along appropriate axis (can be changed in the legend control panel). The predictions are made weekly and each model predicts the following targets for each region (described here):

  1. 4 weeks ahead values of weighted ILI % (wili).
  2. Season onset week. The first week in the range which reaches or exceeds the baseline value for three consecutive weeks
  3. Peak week. Weak with the highest value of wili for this season
  4. Peak wili%. The peak value for wili in this season

More details about these targets can be found here.

Additionally, flusight also displays historical data for reference. While updating, the actual flu data gets backfilled and changes the series . This can be seen below, where the actual data (light green) is lower than the data observed when predictions were made (dark green).

Model Evaluation

Basic information about models' mean performance in the current season is displayed in the Scores tab.

Models are evaluated from the first prediction to the latest prediction provided using mean log score and mean absolute error as mentioned in the FluSight 2016-17 guidelines. Exact details of various scoring metrics used is in the Scoring page.

Clone this wiki locally