The graphs and data forecast depicted on this web site are the output from machine learning forecast models developed at Beyond Ordinary Software Solutions. These models are derived from the Team BOSS XPrize Pandemic Challenge submission and have been updated to include vaccination data and recent reporting trends.
This data and forecasts are not intended to be a public safety recommendation. These models are generated by a computer using data submitted by humans. There are errors in this process, errors that could result in elevated morbidity or even mortality. Listen to your regional health care professionals and their experts. Follow their guidelines. Use the forecast herein as an augmentation tool to further refine your approach to your own personal safety.
Last Updated: 14 June 2023
Forecast Period: 12/01/2022 through 08/31/2023
The models used in this forecast were trained and evaluated using linear PCA. The heatmap of the feature importances can be found here
Almost every political region has a graph associated with it. Each graph shows the reported data in red and the prediction in green. The R2 factor is a measure of variability in the data. An R2 that is 1.0 is a good forecast, whereas an R2 that is negative requires some further interpretation and analysis. The R2 here is the holdout R2, which is the data not used in training that was used to validate the predictive (numerical) ability of the forecast.
Remember this is a forecast, not a prediction. A forecast is a trend, such as things going up or down. A prediction is a number. These models do not accurately predict morbidity and mortality. They have been accurate in forecasting trends and the start of such trends.
The forecast assumes the last reporting day's vaccination trend is the future trend, and that the prior year's same-day mobility data is the same in the future. Not all countries report separation data for each vaccine, so there is an "Other Vaccine" column that represents this mixed bag of vaccines.
Zombies occur in the model. Many countries re-stated historical data by applying corrections (negative reports) in the current reporting. These zombies remain in the model to maintain the historical noise and inaccuracy. The learning model should be able to remove/ignore these events and properly adjust.