**6. Conclusions**

In this study, by using our automatic stress detection system with the use of Empatica-E4 smart-bands, we detected stress levels and suggested appropriate relaxation methods (i.e., traditional or mobile) when high stress levels are experienced. Our stress detection framework is unobtrusive, comfortable and suitable for use in daily life and our relaxation method suggestion system makes its decisions based on the physical activity-related context of a user. To test our system, we collected eight days of data from 16 individuals participating in an EU research project training event. Individuals were exposed to varied stressful and relaxation events (1) training and lectures (mild stress), (2) yoga, mindfulness and mobile mindfulness (PAUSE) (relax) and (3) were required to give a moderated presentation (high stress). The participants were from different countries with diverse cultures.

In addition, 1440 h of mobile data (12 h in a day) were collected during this eight-day event from each participant measuring their stress levels. Data were collected during the training sessions, relaxation events and the moderated presentation and during their free time for 12 h in a day, demonstrating that our study monitored daily life stress. EDA and HR signals were collected to detect physiological stress and a combination of different modalities increased stress detection, performance and provided the most discriminative features. We first applied James Gross ER model in the context of stress managemen<sup>t</sup> and measured the blood pressure during the ER cycle. When the known context was used as the label for stress level detection system, we achieved 98% accuracy for 2-class and 85% accuracy for 3-class. Most of the studies in the literature only detect stress levels of individuals. The participants' stress levels were managed with yoga, mindfulness and a mobile mindfulness application while monitoring their stress levels. We investigated the success of each stress managemen<sup>t</sup> technique by the separability of physiological signals from high-stress sessions. We demonstrated that yoga and traditional mindfulness performed slightly better than the mobile mindfulness application. Furthermore, this study is not without limitations. In order to generalize the conclusions, more experiments based on larger sample groups should be conducted. As future work, we plan to develop personalized perceived stress models by using self-reports and test our system in the wild. Furthermore, attitudes in the psychological field constitute a topic of utmost relevance, which always play an instrumental role in the determination of human behavior [58]. We plan to design a new experiment which accounts for the attitudes of participants towards relaxation methods and their effects on the performance of stress recognition systems.

**Author Contributions:** Y.S.C. is the main editor of this work and made major contributions in data collection, analysis and manuscript writing. H.I.-S. made valuable contributions in both data collection and manuscript writing. She was the yoga and mindfulness instructor in the event and contributed the related sections regarding traditional and mobile methods. She also led the blood pressure measurement efforts before and after relaxation methods. D.E. and N.C. contributed equally to this work in design, implementation, data analysis and writing the manuscript. J.F.-Á., C.R. and G.R. contributed the experiment design and provided valuable insights into both emotion regulation theory. They

also contributed to the related sections in the manuscript. C.E. provided invaluable feedback and technical guidance to interpret the design and the detail of the field study. He also performed comprehensive critical editing to increase the overall quality of the manuscript. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work has been supported by AffecTech: Personal Technologies for Affective Health, Innovative Training Network funded by the H2020 People Programme under Marie Skłodowska-Curie Grant Agreement No. 722022. This work is supported by the Turkish Directorate of Strategy and Budget under the TAM Project number DPT2007K120610.

**Acknowledgments:** We would like to show our gratitude to the Affectech Project for providing us the opportunity for the data collection in the training event and funding the research.

**Conflicts of Interest:** The authors declare no conflict of interest.
