The mindset people have about stress is important to be studied because this core belief, that stress is either enhancing or debilitating, fundamentally alters a person’s physiological and psychological responses to stressors. However, this crucial construct is rarely considered in prior research on momentary stress detection with wearables, leaving two fundamental questions unanswered: can wearable data identify an individual’s stress mindset, and can mindset be leveraged to build better performing stress detection models? To investigate that, we conducted an in-lab study with wearable devices by inducing mental stress in participants (N=23). First, we found that heart rate variability and electrodermal activity features carry signatures of stress mindset. Second, machine learning models can discriminate stress mindset with sensors, achieving AUCs upto 0.88. Finally, a random forest model trained for stress-is-enhancing participants outperformed a one-size-fits-all model (AUC=0.91 vs. 0.78, p<0.05), for the task of stress detection. Our findings show that stress mindset leaves a measurable physiological footprint and that mindset-aware models open the potential for more personalized stress detection and interventions. To support future research, we publicly release the anonymized dataset at https://social-dynamics.net/stress/mindset
ACM CHI Conference on Human Factors in Computing Systems