Revealing Misclassification of Exercise Sessions in Wearable Stress Detection for Automated Insulin Delivery
Sensors · 20.11.2026 · 26 (22) · 7340
DOI: 10.3390/s26227340 ↗Kurz gefasst
Fachartikel in der Zeitschrift Sensors, online erschienen am 20.11.2026. Verfasst von Hamza Abu Owida und 4 weiteren Autor:innen. Die Arbeit nennt 36 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Deanship of Scientific Research and Graduate Studies at King Khalid University.
Abstract (Originalsprache)
Psychological stress tends to raise blood glucose and moderate exercise to lower it, so automated insulin delivery in type 1 diabetes could use wrist-worn stress detection that does not mistake exercise for stress. The only public wristband dataset known to us that combines induced stress with structured exercise (36 participants) reports accuracies of 0.84–0.93, but its evaluation did not keep participants apart, and each class was recorded in a separate session. Under a pre-registered plan, we re-implemented the published evaluation and repeated it on held-out participants, tested whether elapsed time and motion alone separated the classes, and counted stress-detector alarms during exercise. Our re-implementation did not match all published accuracies, held-out evaluation lowered exercise-recognition balanced accuracy by 0.05, and elapsed time alone separated stress from rest above chance. A physiology-only stress detector flagged 96% of exercise-session windows, including 80% of seated windows before exercise began (post hoc). In exploratory analyses, a detector trained with exercise examples rarely flagged exercise within the dataset, although motion features alone did almost as well (post hoc), and it missed a social stress test in an external dataset. Stress detectors for insulin delivery need validation on independent recordings combining strong stress with exercise.
Förderung
- Deanship of Scientific Research and Graduate Studies at King Khalid University
Zitieren
APA
Owida, H. A., Alsarhan, A., Banat, M., Al-Fraihat, D., & Algarni, A. (2026). Revealing Misclassification of Exercise Sessions in Wearable Stress Detection for Automated Insulin Delivery. Sensors, 26(22), 7340. https://doi.org/10.3390/s26227340
Vancouver
Owida HA, Alsarhan A, Banat M, Al-Fraihat D, Algarni A. Revealing Misclassification of Exercise Sessions in Wearable Stress Detection for Automated Insulin Delivery. Sensors. 2026;26(22):7340. doi:10.3390/s26227340
BibTeX
@article{Owida2026,
author = {Owida, Hamza Abu and Alsarhan, Ayoub and Banat, Mohannad and Al-Fraihat, Dimah and Algarni, Abdulmohsen},
title = {Revealing Misclassification of Exercise Sessions in Wearable Stress Detection for Automated Insulin Delivery},
journal = {Sensors},
year = {2026},
volume = {26},
number = {22},
pages = {7340},
doi = {10.3390/s26227340}
}
EUVIMED