Skip to contents

MicroHub Exercises

exercises.Rmd

MicroHub Exercises

Use these exercises to practice moving from a public health forecasting question to a MicroHub forecast output.

Exercise 1: Influenza Hospitalization Forecasts

Scenario

It is January 7, 2026. Influenza-related hospital admissions are a major contributor to the overall burden of influenza in the United States. Accurate forecasts can help public health teams plan staffing, communications, and interventions during the 2025-2026 respiratory virus season, as COVID-19, RSV, and other respiratory pathogens continue to circulate.

The CDC Influenza Division has asked you to produce short-term influenza hospitalization forecasts for the week ending January 10, 2026. You have weekly data from February 2022 through January 3, 2026 for five states: Delaware, Florida, Georgia, Maryland, and North Carolina.

Data

Download exercise-1-flu-hospitalizations.zip from the Downloads page and unzip it before uploading the CSV to MicroHub.

The file includes:

  • date: week-ending date
  • target_group: state
  • population: state population
  • value: influenza-related hospital admissions

Your Task

Using MicroHub, produce four-week influenza hospitalization forecasts for the five states, beginning with the week ending January 10, 2026.

  1. Open MicroHub and upload exercise-1-flu-hospitalizations.csv.
  2. Confirm that MicroHub recognizes date, target_group, population, and value correctly.
  3. Set the forecast date to January 7, 2026.
  4. Set the forecast horizon to 4 weeks.
  5. Select the United States as the country for seasonality.
  6. Run all available forecasting models.
  7. Compare the model plots for the five states.
  8. Create an ensemble forecast from the models you trust most.
  9. Download the forecast output, report, and visualization plots.

Prompts

  • Which states show the clearest increase in recent hospital admissions?
  • Which model appears to respond most strongly to the recent trend?
  • Where do the models disagree the most?
  • Which forecasts would you feel most comfortable sharing with a public health planning team, and why?

Exercise 2: Forecasting with Recent Reporting Anomalies

Scenario

You have weekly target data for three groups, labeled A, B, and C. These groups may represent states, age groups, or another set of populations that a public health team is monitoring.

Cases appear to be increasing in line with expectations for this time of year. In the most recent few weeks, however, the reported counts appear lower than expected. This may reflect true decreases, delayed reporting, incomplete data, or another recent anomaly. Your goal is to decide how much recent data to exclude before producing forecasts and to compare how sensitive the models are to that choice.

Data

Download exercise-2-data.zip from the Downloads page and unzip it before uploading the CSVs to MicroHub.

The ZIP includes:

  • exercise-2-2025-12-15.csv
  • exercise-2-2024-02-01.csv

Each file includes:

  • date: week-ending date
  • target_group: group label
  • population: group population
  • value: reported cases

Your Task

Start with exercise-2-2025-12-15.csv.

  1. Open MicroHub, upload the file, and explore the visualizations for all three target groups.
  2. Confirm the forecast date and select the seasonal zone that best matches the data.
  3. Try several data-drop settings within your group. Decide how many recent weeks to drop before forecasting.
  4. Run each individual forecast model using the default parameterization.
  5. Change selected model parameters and compare how the forecasts respond.

Then repeat the same workflow with exercise-2-2024-02-01.csv. As you work, consider what part of the outbreak season the dataset represents and how that timing might affect forecast behavior and reliability.

Prompts

  • For each dataset, how many weeks of data did you drop, and why?
  • How did the forecasts change when you changed the number of dropped weeks?
  • Did some models respond more strongly than others when you dropped additional weeks?
  • What did you observe when varying model parameters?
  • Did some parameters have more impact on model predictions than others?
  • How did the timing of the exercise-2-2024-02-01.csv dataset affect your confidence in the forecasts?