Executive Summary
Beginning in November 25, 2024, AI4Casting Hub will begin accepting submissions for two forecasting challenge initiatives:
- Hospital Bed Occupancy Forecasting for COVID-19, RSV, and Flu in Ontario and Its Public Health Regions.
- Respiratory Virus Detection Surveillance System’s Percentage of Positive Laboratory Detections Data Forecasting in Canada and it’s regions for SARS-CoV-2, Influenza and RSV.
For the most up to date and complete guidelines for this submitting this season’s forecasts please see detailed information provided in the README documentation throughout the AI4Casting Hub’s respective GitHub repository (https://github.com/ai4castinghub). Please also see attached example submission file for formatting reference, particularly regarding new optional forecast targets.
1) Hospital Bed Occupancy Forecasting for COVID-19, RSV, and Flu in Ontario and Its Public Health Regions
Challenge Period
The forecast period begins on November 25, 2024, and will run until May 31, 2025. Participants are asked to submit weekly forecasts every Saturday by 11 PM Eastern Time.
Data Source
We use data from Public Health Ontario’s Ontario Respiratory Virus Tool, which provides comprehensive epidemiological insights on respiratory viruses, including hospital capacity monitoring. This data helps set our target predictions for hospital bed occupancy counts for COVID-19, RSV, and influenza. For detailed information on target data, click here.
Submission Guidelines
Details on formatting and the submission procedure are provided here: https://github.com/ai4castinghub/hospitalization-forecast/tree/main/model-output
Forecast submissions are organized by a reference date—the Saturday following each due date, which marks the end of the epidemiological week (EW) from Sunday to Saturday. All submissions should be formatted as .csv files.
Prediction Targets
Each weekly forecast should include predictions for:
- Weekly COVID-19 Hospital Bed Occupancy Count (wk inc covid hosp)
- Weekly Influenza Hospital Bed Occupancy Count (wk inc flu hosp)
- Weekly RSV Hospital Bed Occupancy Count (wk inc rsv hosp)
While teams are encouraged to forecast for all three targets, submissions are not required for every target.
Prediction Locations
Forecasts should cover Ontario at the provincial level and its 6 Public Health Regions. Teams are asked to submit nowcasts and predictions for each location, though predictions for every location are optional.
Prediction Horizons
Forecasts should cover:
- The prior week (horizon -1)
- The current week (horizon 0)
- The next three weeks (horizons 1, 2, and 3)
Forecasts for the previous week (horizon -1) are optional and will not be included in scoring.
Prediction Output
Each submission file should include an output_type column set to “quantile” to represent probabilistic forecasts for hospital bed occupancy. Teams must provide forecasts for the following 7 quantiles: 0.025, 0.1, 0.25, 0.5, 0.75, 0.9, and 0.975.
In the value column, teams should enter non-negative numbers representing the quantile prediction for each row. These values correspond to the inverse cumulative distribution function (CDF) for each target, location, and quantile level.
Example
Below is an example submission for flu hospital bed occupancy in the Central region for the week ending October 12, 2024. Predictions cover quantiles from the 2.5th to the 97.5th percentiles. Here, horizon -1 indicates a hindcast for the week before the reference date of October 19, 2024.
2) Respiratory Virus Detection Surveillance System’s Percentage of Positive Laboratory Detections Data Forecasting in Canada and it’s regions for SARS-CoV-2, Influenza and RSV.
Challenge Period
The forecast period begins on November 25, 2024, and will run until May 31, 2025. Participants are asked to submit weekly forecasts every Saturday by 11 PM Eastern Time.
Data Source
We use data from Respiratory Virus Detection Surveillance System, which collects data from select laboratories across Canada on the number of tests performed and the number of positive tests for up to 8 respiratory viruses. This data helps set our target predictions for percentage positive lab detection for SARS-CoV-2, RSV, and influenza. For detailed information on target data, click here.
Submission Guidelines
Details on formatting and the submission procedure are provided here: https://github.com/ai4castinghub/rvdss-forecast/tree/main/model-output
Forecast submissions are organized by a reference date—the Saturday following each due date, which marks the end of the epidemiological week (EW) from Sunday to Saturday. All submissions should be formatted as .csv files.
Prediction Targets
Each weekly forecast should include predictions for:
- Percentage of Weekly SARS-CoV-2 Positive Lab Detections (pct wk covid lab det)
- Percentage of Weekly Influenza Positive Lab Detections (pct wk flu lab det)
- Percentage of Weekly RSV Positive Lab Detections (pct wk rsv lab det)
While teams are encouraged to forecast for all three targets, submissions are not required for every target.
Prediction Locations
Forecasts should cover Canada at the national level, at the provincial level (e.g., Ontario, British Columbia), and at the regional level (e.g., Atlantic, Prairies). Teams are asked to submit nowcasts and predictions for each location, though predictions for every location are optional.
Prediction Horizons
Forecasts should cover:
- The prior week (horizon -1)
- The current week (horizon 0)
- The next three weeks (horizons 1, 2, and 3)
Forecasts for the previous week (horizon -1) are optional and will not be included in scoring.
Prediction Output
Each submission file should include an output_type column set to “quantile” to represent probabilistic forecasts for hospital bed occupancy. Teams must provide forecasts for the following 7 quantiles: 0.025, 0.1, 0.25, 0.5, 0.75, 0.9, and 0.975.
In the value column, teams should enter non-negative numbers representing the quantile prediction for each row. These values correspond to the inverse cumulative distribution function (CDF) for each target, location, and quantile level.
Example
Below is an example submission for percentage positive flu lab detections in the Ontario province for the week ending October 12, 2024. Predictions cover quantiles from the 2.5th to the 97.5th percentiles. Here, horizon -1 indicates a hindcast for the week before the reference date of October 19, 2024.
Thank you for your participation in the AI4Casting Hub Challenges.
Please feel free to email 4castinghub@uoguelph.ca with any questions!