STREL - Naturalistic Dataset and Methods for Studying Mental Stress and Relaxation Patterns in Critical Leading Roles

We investigate mental stress and relaxation patterns in professionals occupying leadership roles in emergency care and special police units. A key finding is that on days that involve critical missions, these individuals experience negative mental stress (i.e., distress) that escalates as the day unfolds. In contrast, on non-leadership workdays, mental distress remains relatively stable, while on non-workdays, participants exhibit positive mental stress (i.e., eustress) that subsides over time, facilitating relaxation. These findings stem from a four-day naturalistic study of $n=24$ professionals, during which we collected physiological, mobility, and psychometric data. Using participant debriefings and sensor-based validation triggers (e.g., GPS, cadence), we labeled activities in 5-minute intervals and focused our analysis on sedentary periods. We defined stress during these periods as a normalized heart rate that exceeds two standard deviations above a personalized baseline, thus isolating mental stress uncontaminated by physical exertion. A logistic regression model based on this stress labeling method yielded results largely consistent with those obtained from Kubios' SNS Index, reinforcing its validity. In cases of disagreement, our method aligned better with participant reports and established literature, highlighting advantages in interpretability and specificity. Overall, our work makes three contributions: (a) to affective science, by quantifying the mentally stressful nature of leadership in high-stakes environments; (b) to affective computing, by proposing a wearable compatible method for estimating mental stress during sedentary activity in the wild; (c) to data science, by introducing a well-annotated, multimodal dataset suitable for machine learning benchmarking in stress detection.