Modern health apps are moving beyond simply recording past behavior and are becoming proactive partners that can anticipate physical needs before they become problems. While diet and medication are central to managing Type 2 Diabetes, physical activity also plays a critical role in regulating glucose levels.
In this talk, I present findings from a study of 103 older adults that show why commonly used metrics, such as step counts, are insufficient. Generic recommendations often fail because they ignore the unique and personal rhythms of how people actually move. We introduce a new concept called Activity Fingerprints, which are individualized movement signatures derived by applying Fast Fourier Transform to posture signals collected from wearable sensors.
Our results show that these rhythms function as leading indicators of glucose change. Using Granger causality analysis, we identify a predictive horizon of approximately 40 minutes. An XGBoost model trained on Activity Fingerprints achieves a prediction error of 17.99 mg/dL, compared with 144.50 mg/dL for a standard last value carried forward baseline.
These findings suggest that posture dynamics can provide a non-invasive and personalized signal for timely intervention, enabling people to take action before adverse glucose events occur.
Veronika Bogina is a researcher at the University of Haifa specializing in recommender systems, user modeling, and AI for healthcare. Her work bridges machine learning, causality, and personalization, with a strong focus on real-world impact across academia and industry, including experience at IBM.She often finds her best research ideas on long nature hikes, ideally followed by excellent food.
Keynote session: Hadas Grossmon Ella
Break
Lightning talks session
Roundtable closing
Talk by Hila Paz
Talk by Dr. Moran Mizrahi
Closing remarks
End
