MicroHub Forecasting empowers public health officials worldwide with timely, actionable insights to support effective responses to respiratory illness outbreaks — wherever they occur.
The Visualization
The visualization shows the model-specific forecasts for each target group.
Black dots indicate the recent season’s data that was used by the forecast model, blue dots indicate the recent data (if any) that was excluded from the forecast model, and the red dots show the data from the previous year.
For the forecasts, the black line indicates the expected path of the epidemic, with the 50% and 95% prediction intervals indicated by the dark and light grey shaded regions respectively. We use these plots during our weekly forecasts to identify potential issues or parameters that may need to be tuned.
For example, if the most recent data point is significantly lower than the other recent trends and is out of touch with expected seasonality, it provides evidence that we may want to drop that data point from being included in the forecast model because it will likely have significant backfill in future weeks.
Run All Models Run every model with default model settings using the shared data settings above.
Download Data Template Download the template and replace the example data with your target data. See instructions for using the data template.
The Regular Baseline assumes the next few weeks will look broadly like the usual week-to-week changes seen in your historical data. It is a simple benchmark that helps show whether a more complex model is really adding value. More details
Run the model to generate forecast plots. The plot area stays fixed so results remain easy to compare across tabs.
Download Regular Baseline Plot (.png)The Seasonal Baseline focuses on the repeating shape of past respiratory seasons and projects that typical seasonal pattern forward. It is useful when timing and rise-and-fall behavior tend to be fairly consistent from year to year. More details
Run the model to generate forecast plots. The plot area stays fixed so results remain easy to compare across tabs.
Download Seasonal Baseline Plot (.png)The Opt Baseline is a faster-reacting version of the baseline that pays more attention to the most recent part of the series instead of the full history. It is often useful when conditions are changing and older seasons may be less informative. More details
Run the model to generate forecast plots. The plot area stays fixed so results remain easy to compare across tabs.
Download Opt Baseline Plot (.png)INFLAenza combines recurring seasonal patterns with the most recent changes in the data to estimate where the series is headed next. It is a flexible forecasting model that works well when both long-term seasonality and short-term movement matter. More details
Run the model to generate forecast plots. The plot area stays fixed so results remain easy to compare across tabs.
Download INFLAenza Plot (.png)Copycat looks for past seasons that behaved like the last few weeks of your current data, then projects those historical trajectories forward as possible futures. It is essentially a pattern-matching model built around the question: what happened next the last time things looked like this? More details
Run the model to generate forecast plots. The plot area stays fixed so results remain easy to compare across tabs.
Download Copycat Plot (.png)newGBQR is a year-round version of GBQR that removes fixed in-season windows, learns peak timing empirically from the uploaded data, and adds cyclic week-of-year features for forecasts in any week. More details
Run the model to generate forecast plots. The plot area stays fixed so results remain easy to compare across tabs.
Download newGBQR Plot (.png)The Ensemble combines selected models into one shared forecast by taking the median prediction across them. It is useful when you want a more stable result that is less sensitive to the quirks of any single model. More details
Run the model to generate forecast plots. The plot area stays fixed so results remain easy to compare across tabs.
Download Ensemble Plot (.png)Download the forecasts you have generated as a single standardized results file. Use the preview below to quickly check the output before exporting.
Choose which forecast models to include, then export the filtered results as one CSV file.
Export every generated app plot as a multi-page PDF, one model per page.
Download Plots (.pdf)Preview the combined output table before downloading.
CalCopycat starts with the same historical pattern-matching forecast as Copycat, then calibrates its uncertainty using leave-one-out historical forecast errors from comparable weeks in the time series. That makes its intervals more grounded in observed forecast error. More details
Run the model to generate forecast plots. The plot area stays fixed so results remain easy to compare across tabs.
Download CalCopycat Plot (.png)The FourCAT (Fourier-augmented Calendar Aware Transformer) model is a deep learning model for probabilistic epidemic forecasting. It uses a Transformer encoder with Fourier-based epiweek embeddings to capture seasonal patterns, and outputs calibrated quantile forecasts via a pinball loss objective.
The model is trained across multiple random seeds and forecasts are ensembled by averaging quantile predictions across seeds to improve robustness.
No additional settings are required. FourCAT uses the forecast date, data to drop, and forecast horizon configured in the Data Upload & Settings tab.