When Time Disappears: Uncovering Stress in an Analog Underground Mission

Abstract: Stress is a central component of human adaptation, particularly in isolated confined extreme (ICE) environments. In such settings, stressors may impair cognitive performance, emotional regulation, decision-making processes, and overall psychological and physical well-being. ICE environments also provide unique opportunities to investigate the boundaries of human adaptability, particularly when they involve temporal and social isolation. This study examined whether heart rate variability (HRV)-derived features could predict perceived stress in a controlled laboratory setting, as well as whether these models could be applied to an analogue ICE experiment without access to self-reported stress assessment. Supervised classification models were trained on the SPACE dataset using HRV features and ratings from the Visual Analogue Scale for Stress (VAS-S). The best-performing models were then applied, without retraining, to physiological data collected during the 15-day DeepTime II cave isolation mission. In the absence of subjective labels, the validity was examined using Baevsky’s Stress Index (BSI) as an autonomic reference marker. There was substantial variation in HRV-based models between individuals in the SPACE dataset, and models performed only marginally better than chance at differentiating stress from no-stress conditions. Despite substantial class overlap, predicted stress proxies exhibited descriptive differences in BSI across predicted categories, with higher predicted classes tending to show higher autonomic strain. In the absence of subjective assessments, cardiac autonomic indicators alone provide limited inference of perceived stress, particularly when models are applied to fundamentally different contexts. These findings highlight the constraints of generalized HRV-based stress modelling and support the need for individualized and multimodal approaches in ICE environments.

Keywords: Isolated Confined Extreme (ICE), Heart Rate Variability, Perceived Stress, Supervised Machine Learning, Model Transferability.

DOI: 10.54941/ahfe1007176

Cite this paper:

​