Real-Time Detection of Student Stress During Timed Exams

• Trey Prospero, Mathematics, University of Delaware

• Matthew Mauriello, Computer science, University of Delaware

Stress is a common but poorly understood factor in how students perform on high-stakes, timed assessments. Prior work on stress in programming contexts often relies on retrospective self-report, which cannot capture the moment-to-moment physiological and behavioral changes that accompany acute stress during a task. This project addresses that gap by designing a study that observes physiological and behavioral indicators of stress during a real, timed CS1 exam and tests whether stress-detection software can accurately identify and respond to stress in real time. Our research questions ask how physiological and behavioral signals (e.g., heart rate/HRV, facial expressions, and typing/mouse behavior) relate to students’ stress and emotion, how accurately automated software can detect genuine moments of stress, and whether visually distinct stress levels can be identified that reflect the expectation that moderate levels of stress are associated with optimal performance, whereas both low and excessive stress are associated with poorer performance. Approximately 15 undergraduate students will complete a 90-minute, proctored, computer-based exam while wearing a heart-rate monitor and being recorded by webcam and additional cameras for facial, postural, and keystroke/mouse analysis; all participants will complete pre- and post-exam surveys. Physiological and behavioral logs will be analyzed using statistical and machine-learning methods to examine alignment between continuous signals and self-reported stress. Findings from this feasibility work will inform whether automated detection meaningfully tracks subjective stress experience.