Determining Prognostic Physiological Attributes of Stress Using Wearable Sensors

Egilmez, Begum

doi:https://doi.org/10.21985/n2-3t70-zp88

Work

Determining Prognostic Physiological Attributes of Stress Using Wearable Sensors

Public

Download PDF

This thesis explains the works that have been completed towards the Ph.D. thesis of the author and discusses the conclusions derived from the results, as well as what future holds for stress-related mobile health research. Main focus of the thesis is use of wearable sensors to understand physiological manifestation of stress in two chosen populations that are prone to high levels of stress in daily life. This physiological manifestation is used for both in-the-moment stress detection and future stress prediction. There are previous works in the literature that approached the issue of stress detection from physiological point of view with use of non-invasive passive wearable sensors [54, 63, 116, 51], however, neither of them tackled the issue specifically for college students or pregnant women under prenatal stress. In this thesis, author is focusing on these two chosen populations and sheds light on their unique stress cases and which features and models dene these stress cases. The first work in this field that is completed by the author, have helped with identifying predictive features of in-the-moment stress of college students in controlled environments. Based on the results gathered from this study, controlled environment study design is improved, device selection is updated and a longer in-field study is conducted with college students in the next study. Results of this work showed stress detection models built from self-reports collected in controlled environment experiments are not sufficient for stress detection in-field. Personalized models built from in-field data perform the best in in-field in-the-moment stress detection. Also, models using ECG data rather than heart rate and galvanic skin response perform better with 0.71 F1-score. In the final work, prenatal stress of pregnant women is investigated to build self-reported stress prediction for next-day stress. Results of this study showed that in chronic stress cases such as prenatal stress, there is a remembrance factor that causes increase in the self-reports but does not show manifestation in the physiology like acute stress episodes. Thus, physiological data, specically electrocardiography (ECG) data which is shown to be an effective estimator of in-the-moment stress in the previous work, is not sufficient by itself to predict next-day self-reports. However, if an effective prenatal stress intervention course is also taking place, information about the schedule of this intervention can provide signicant insights on next-day stress levels, also confirming the effectiveness of the intervention technique used in the study.

Creator