Wearables allow us to measure our bodily and mental state in real life, but the interpretation of these signals is not always straightforward. In order to nevertheless develop a stress indicator for use in wearables, we focused on skin conductance, since it almost instantaneously reacts to psychological stress. As a ground truth, we also looked at the stress hormone cortisol, which is known to peak about 20-30 min later. We modeled the relation between the two as a convolution of the height of the skin conductance peaks with the cortisol stress response curve. Thus we obtained a skin conductance-derived estimate of stress-induced cortisol, and we implemented this in an algorithm.

In order to validate this model we compared the skin-conductance derived stress estimates with cortisol levels actually measured in saliva samples in a group of 46 participants. They performed a set of stressful, boring, and performance tasks, while we measured their skin conductance continuously and took salivary cortisol samples at several moments in time. Most participants indeed showed an increase in salivary cortisol after the stressful task, but a sizeable group did not. For both groups, we found that the correlation between our algorithm’s stress estimates and the measured salivary cortisol was substantial. Therefore, these results support the use of our algorithm as a stress indicator.