**3. Related Work**

Researchers have created the ability to detect stress in laboratory environments with medical-grade devices [25–28]; smartwatches and smart bands started to be used for stress level detection studies [29– 31]. These devices provide high comfort and rich functionality for the users, but their stress detection accuracies are lower than medical-grade devices due to low signal quality and difficulty obtaining data in intense physical activity. If data are collected for long periods, researchers have shown that their detection performance improves [32]. During movement periods, the signal can be lost (gap in the data) or artifacts might be generated. Stress level detection accuracies for 2-classes by using these devices are around 70% [29,30,33,34].

After detecting the stress level of individuals, researchers should recover from the stressed state to the baseline state. To the best of our knowledge, there are very few studies that combine automatic stress detection (using physiological data) with recommended appropriate stress managemen<sup>t</sup> techniques. Ahani et al. [35] examined the physiological effect of mindfulness. They used the Biosemi device which acquires electroencephalogram (EEG) and respiration signals. They successfully distinguished control (non-meditative state) and meditation states with machine learning algorithms. Karydis et al. [36] identified the post-meditation perceptual states by using a wearable EEG measurement device (Muse headband). Mason et al. [37] examined the effect of yoga on physiological signals. They used PortaPres Digital Plethtsmograph for measuring blood pressure and respiration signals. They also showed the positive effect of yoga by using these signals. A further study validated the positive effect of yoga with physiological signals; researchers monitored breathing and heart rate pulse with a piezoelectric belt and a pulse sensor [21]. They demonstrated the effectiveness of different yogic breathing patterns to help participants relax. There are also several studies showing the effectiveness of mobile mindfulness apps by using physiological signals [20,38,39]. Svetlov et al. [20] monitored the heart rate variability (HRV), electrodermal activity (EDA), Salivary alpha-amylase (sAA) and EEG values. In other studies, EEG and respiration signals were also used for validating the effect of mobile mindfulness apps [38,39]. When the literature is examined, it could be observed that the effect of ancient relaxation methods and mobile mindfulness methods are examined separately in different studies. Ancient methods generally require out of office environments that are not suitable for most of the population, since, in the modern age, people started to spend more time in office-like environments. On the other hand, some smartphone applications such as Pause, HeartMath and Calm do not require extra hardware or equipment and be applicable in office environments. Hence, an ideal solution depends on the context of individuals. A system that monitors stress levels, analyzes the context of individuals and suggests an appropriate relaxation method in the case of high stress will benefit society. Furthermore, mobile methods along with the ancient techniques should be applied in stressful real-life events and their effectiveness should be compared by investigating physiological signals. When the literature is examined, there is not any study comparing the performance of these methods in real-life events (see Table 1). Another important finding is that these methods should be compared with unobtrusive wearable devices so that they could be used for a biofeedback system in daily lives. Individuals may be reluctant to use a system with cables, electrodes and boards in their daily life. Therefore, a comparison of

different states with such systems could not be used in daily life. There is clearly a need for a suggestion and comparison of ancient and mobile meditation methods by using algorithms that could run on unobtrusive devices. An ideal system should detect high stress levels, sugges<sup>t</sup> relaxation methods and control whether users are doing these exercises right or not with unobtrusive devices. Our algorithm is suitable to be embedded in such daily life applicable systems that use physiological signals such as skin temperature (ST), HRV, EDA and accelerometer (ACC). In this paper, we present the findings of our pilot study that tested the use of our algorithm during general daily activities, stress reduction activities and a stressful event.


**Table 1.** Comparison of our work with the studies applying different types of meditation techniques for stress managemen<sup>t</sup> in the literature.
