Severity-Weighted Hazard Coverage Gaps in Wearable Stress Detection for Mining and Construction: A Simulation-Based Quantification Framework

This is the authors' abstract. We don't add key points for this paper.

Wearable systems promise real-time detection of workers’ acute physiological stress in underground mining and construction, yet whether existing detectors are trained on the stressors that injure and kill workers has not been examined. This simulation-based methodological study introduces a framework that (1) harmonizes U.S. Mine Safety and Health Administration and Occupational Safety and Health Administration incident narratives into a cross-industry hazard taxonomy and a severity-weighted acute stressor burden, (2) benchmarks five shallow classifiers on heart rate variability and electrodermal features under leave-one-subject-out and leave-one-dataset-out protocols, and (3) composes the two through a severity-weighted Hazard–Stressor Coverage Gap Index built on an explicitly documented assumed coverage matrix. The framework is demonstrated on literature-calibrated simulated inputs that reproduce the documented structure of 13,000 narratives and four public wearable datasets (WESAD, SWELL-KW, Nurse-Stress, and Exam-Stress). In this instantiation, the most lethal hazards were not the most frequent, detector performance degraded from within- to cross-dataset evaluation mainly in threshold-dependent metrics, and existing benchmark paradigms represented roughly one-fifth of the severity-weighted stressor burden (0.17 to 0.35 under ±0.20 perturbation of every coverage entry). Sudden physical threat and near-miss stress accounted for about two-thirds of the uncovered burden and ranked first under every perturbation tested. The reported numbers characterize the pipeline on simulated inputs. The released pipeline runs unchanged on the public files. The contribution is a reproducible, severity-weighted method for prioritizing where new physiological data collection would most improve worker safety monitoring.