| dc.description.abstract | For some, public speaking can cause heightened moments of stress while giving a speech or presentation. These moments are quantifiable through one’s physiology and vocal characteristics, measurable through sensor-enabled smart technology. Through these measurements, we can assess the current state of the individual to determine opportune moments to deliver interventions that alleviate symptoms of stressful moments.
Recent work in wrist-worn vibrotactile biofeedback suggests that it is a promising intervention towards reducing state-based anxiety for public speaking. However, since the vibrotactile stimulus is delivered constantly, adaptation could risk diminishing relieving effects. Therefore, we administer vibrotactile biofeedback as a just-in-time adaptive intervention during in-the-moment heightened levels of stress. We evaluate two types of vibrotactile feedback delivery mechanisms in a between-subjects design – one that delivers stimulus randomly and one that delivers stimulus during moments of heightened physiological reactivity, as determined by changes in electrodermal activity. The results from these interventions indicate that vibrotactile biofeedback administered during high physiological arousal appears to improve stress-related measures early on, but these effects diminish over time. However, we also observe no significant differences in self-reported state anxiety scores between experiment groups.
In the latter half of this thesis, we will explore methods for personalizing machine learning models that detect the onset of heightened moments of stress in real-time. Results indicate that baseline-norming, fine-tuning on participant-specific data, and providing individual-specific trait information are all helpful techniques for improving stress detection performance. | |
