More Models Improve COVID-19 Ensemble Forecasts

During infectious disease outbreaks, short-term forecasts are critical for decision-makers and the public, yet it has remained unclear how the number of contributing models or their specific selection criteria influence the final aggregate performance.

The team investigated whether increasing the pool of voluntary model contributions leads to better outcomes, given that modelling teams often face competing priorities. They also examined if selecting models based on past performance or other characteristics yields advantages over including all available models equally. The central question was whether optimizing the composition of the ensemble offers more benefits than simply increasing participation to moderate levels.

The analysis revealed that including more models in the ensemble both improved and stabilized the aggregate performance. The study found that this approach is particularly relevant for platforms relying on voluntary contributions, where the number of available models can fluctuate based on the priorities of the participating teams.

Conversely, the researchers found that selecting for better component models did not yield any particular advantage. Whether models were chosen based on their historical accuracy or other qualitative characteristics, the performance of the resulting ensemble did not significantly differ from that of an ensemble that included all models equally.

Furthermore, the study examined the role of diversity among the models. The researchers measured diversity both numerically and qualitatively to see if a wider range of modelling approaches contributed to better ensemble performance. However, they found that diversity did not have a clear impact on the overall accuracy or consistency of the forecasts.

The findings offer practical guidance for organizers of collaborative prediction platforms during outbreaks. The results suggest that there are more obvious gains to be made from increasing participation to moderate levels than from attempting to optimize the specific component models. By encouraging a broader range of teams to contribute, platforms can enhance the stability and accuracy of their combined real-time forecasts without needing to implement complex selection criteria.

Subject of Research: Impact of ensemble size and composition on COVID-19 forecast accuracy

Article Title: The influence of ensemble size and composition on the performance of combined real-time COVID-19 forecasts

Article References: Becker, F., Sherratt, K., Bosse, N. I., & Funk, S. (2026). The influence of ensemble size and composition on the performance of combined real-time COVID-19 forecasts. PLOS Computational Biology, 22(11), e1014908. https://doi.org/10.1371/journal.pcbi.1014908

Image Credits: AI Generated

DOI: 10.1371/journal.pcbi.1014908

Keywords: COVID-19 forecasting, ensemble methods, prediction platforms, epidemiology, model selection, European Forecast Hub, public health, statistical modeling, influence, ensemble, size, composition