Abstract:
In the last decades, most of the diseases in modern society are caused by stress. This is the reason researchers want to detect and alleviate stress in daily life as early as possible. With the advance of technology, smartphones, smartbands, watches have become integral items of our daily lives. The research question that whether detecting stress with these widely used wearable devices is possible has arisen. The research has started in laboratory environments and recently a number of works have taken a step outside the laboratory to real life. In this thesis, we employed two different case studies. First, we collected 339 hours of physiological data and 7119 workload survey questions from 17 participants in their real-life environments with Samsung Gear S2 smartwatch. The duration of this experiment for the participants was a month on average. Heart rate variability and accelerometer features are used to evaluate the level of stress. Second, we conducted a context-driven stress measurement experiment, we collected 672 hours (in 9 days) of physiological data from 21 participants of an algorithmic competition event. This event has free, lecture and contest sessions. By using heart rate, skin conductance, and accelerometer signals, we achieved approximately 98% accuracy of discriminating contest stress, the cognitive load (lecture) and relaxed activities by using machine learning methods.