Development of an early alert model for pandemic situations in Germany

Abstract

The COVID-19 pandemic has pointed out the need for new technical approaches to increase the preparedness of healthcare systems. One important measure is to develop innovative early warning systems. Along those lines we developed a machine learning (ML) approach using Google Trends and Twitter data for Germany. As a use case we evaluated our models using COVID-19 surveillance data between March 2020 and June 2022. In conclusion we found that a long-short-term memory (LSTM) jointly trained on COVID-19 symptoms related Google Trends and Twitter data was able to accurately forecast up-trends in confirmed cases and hospitalization rates 14 days ahead. In both cases, F1 scores were above 98% and 97%, respectively, hence demonstrating the potential of using digital traces for building an early alert system for pandemics in Germany.

A dynamic ensemble model for short-term forecasting in pandemic situations

“…As metadata, we incorporated data from Google Trends following Wang et al [24]. First, we identified the 20 top symptoms of COVID-19 which were used as search terms in Google Trends.…”

“…Our above-described modeling approaches used only surveillance data, their forecasts, and estimated prediction performances as input for the base models and meta-model, respectively. In our previous publication, we showed that social media data is not only correlated with surveillance data but can also be used to forecast up-and downtrends of pandemic waves [24]. Therefore, we wanted to test if the inclusion of social media data or further metadata could improve the prediction performance of the meta-model.…”

“…The metamodel was then trained in the same way as before, but now the weights of the LSTM were also updated according to the weighted MSE loss between the output and the real values. Due to the high computational burden and to be consistent with Wang et al [24], we evaluated this approach on German surveillance data only.…”