Phenotyping stress in virtual reality environment

Abstract

This research investigates the relationship between self-reported stress levels and physiological stress indicators within a simulated office environment in virtual reality (VR), with the goal of developing machine learning classifiers to automatically predict stress states. As VR proliferates across industries, insights into mitigating potential negative impacts like VR induced symptoms and stress bear importance from productivity, ethical, and employee wellbeing perspectives. This study exposes participants to a workplace-relevant VR stress exposure paradigm and collects multi-dimensional profiles of stress reactions, combining self-reported anxiety levels from a State-Trait Anxiety Inventory (STAI) questionnaire with facial expressions tracked via a VR headset. The study adopts a two-category experimental framework with a stress condition (C1) designed to deliberately induce stress, and a neutral condition (C2) with non-stressful tasks. Multimodal data is captured throughout, including facial expressions, headset motions, task performance, and self-reports. A Random Forest classifier is developed to classify stress levels, achieving 62.14% accuracy. Correlations emerged between physiological markers and self-reported stress levels, demonstrating feasibility of machine learning for automatic stress recognition in VR environments. The results advance understanding of VR’s potential for developing accurate computational models for stress detection. By blending real and virtual experiences, the research pioneers new methodologies in immersive, controlled environments. Limitations of dataset size and noise are noted. Future work should expand datasets, refine data collection/analysis, explore advanced machine learning, develop stress interventions in VR, and investigate their long-term impacts. This timely investigation makes important contributions towards automated stress classification using psychological and physiological cues captured in virtual reality settings. The insights contribute significantly to research on affective computing, emotion recognition, and technological integrations for holistic stress analysis. Keywords: virtual reality, stress detection, stress assessment, stress phenotyping