Hovsepian, K., al’Absi, M., Ertin, E., Kamarck, T., Nakajima, M., & Kumar, S. (2015). Stress: Towards a Gold Standard for Continuous Stress Assessment in the Mobile Environment. Proc ACM Int Conf Ubiquitous Comput, 493-504.
This study describes the development and testing of cStress, an algorithm to predict stress based on data collected from wearable heart rate and respiration sensors. Investigators collected sensor data (using a chest band sensor suite called Autosense) from subjects who participated in standard laboratory stress tasks (public speaking, mental arithmetic, cold-pressor) and a one-week field study. Ground truth was assessed by self-report of stress with ecological momentary assessments via mobile phones. cStress yielded 88.6% true positive predication of self-reported stress in the laboratory (4.7% false positive), and 72% accuracy in the field. Detailed training, testing and data processing procedures are described. Prediction of stress could be improved by collection of additional data, including location, what people are seeing, and how people are using technology and social media.