Abstract: Stress is the root cause of many diseases. Being able to monitor when and why a person is stressed could inform personal stress management as well as interventions when necessary. In this thesis, I present StressAware, an application on the Amulet wearable platform to measure the stress levels of individuals continuously and in real time. The app implements a stress detection model, continuously streams heart rate data from a commercial heart-rate monitor such as a Zephyr and Polar H7, classifies the stress level of an individual, logs the stress level and then displays it as a graph on the screen. I developed a stress detection model using a Linear Support Vector Machine. I trained my classifiers using data from 3 sources: PhysioNet, a public database with various physiological data, a field study, where subjects went about their normal daily activities and a lab study in a controlled environment, where subjects were exposed to various stressors. I used 73 data segments of stress data obtained from PhysioNet, 120 data segments from the field study, and 14 data segments from the lab study. I extracted 14 heart rate and heart rate variability features. With 10-fold cross validation for Radial Basis Function (RBF) SVM, I obtained an accuracy of 94.5% for the PhysioNet dataset and 100% for the field study dataset. And for the lab study, I obtained an accuracy of 64.29% with leave-one-out cross-validation. Testing the StressAware app revealed a projected battery life of up to 12 days before needing to recharge. Also, the usability feedback from subjects showed that the Amulet and Zephyr have a potential to be used by people for monitoring their stress levels. The results are promising, indicating that the app may be used for stress detection, and eventually for the development of stress-related intervention that could improve the health of individuals.
Copyright © 2016 by the author.The copy made available here is the authors' version; for a definitive copy see the publisher's version described above.