Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

BP: Fachverband Biologische Physik

BP 14: Poster Session II

BP 14.69: Poster

Dienstag, 10. März 2026, 18:00–21:00, P2

Dynamic Health Monitoring: Predicting COVID-19 with Wearable Sensor Data and catch22 Features — •Paul Buttkus and Dirk Brockmann — Technical University Dresden (SynoSys), Dresden, Germany

In the initial stages of a pandemic, when intervention leverage is highest, controlling infectious-disease spread hinges on timely detection of emerging cases. The rapid, global transmission of COVID-19 highlighted the need for scalable sensing tools that can pick up early physiological signatures of infection. Using data from the Corona Data Donation Project, which provides resting heart rate, step count, and sleep duration time series from over 120,000 voluntary participants in uncontrolled, real-world settings, we trained regression models to classify COVID-19 test results. Based solely on daily aggregated features, this model achieved an above random guessing success rate, revealing a non-trivial signal despite coarse temporal resolution and strong noise. To better highlight the underlying dynamics, we employ the catch22 (22 CAnonical Time-series Characteristics) feature set to map raw sensor data to a compact set of interpretable descriptors, and additionally extend our framework to higher-temporal-resolution data to incorporate periodicity metrics (e.g., circadian modulation of heart rate and activity) that are lost under daily aggregation. We show which dynamical features are most informative for distinguishing COVID-19-positive from negative individuals and discuss how this framework could turn large-scale wearable data into a real-time surveillance tool for public health.

Keywords: Wearable sensors; COVID-19 detection; Time-series analysis; Predictive Modeling; Public health