Answers are grounded in this paper and cite their sources.
Workers’ ability to detect hazard cues is essential for construction safety, yet continuous assessment during hot-weather work remains challenging. This study investigated whether wearable-sensor-derived physical workload and individual characteristics could predict declines in hazard-cue information acquisition after work. Thirty-one male construction workers contributed 62 morning and afternoon work-period records. Wrist-worn triaxial accelerometers provided mean metabolic equivalents (METs), which were combined with body mass index, age, work experience, physical fitness, and work period. Decline was defined as a reduction in accuracy on a brief visual hazard-cue information-acquisition task administered before and after work. Four classifiers, including L2-regularized logistic regression, elastic net logistic regression, a shallow random forest, and a constrained gradient-boosting tree, were evaluated using nested leave-one-worker-out cross-validation, with hyperparameters selected by the Brier score. Thirty-four records showed declining hit rate. Elastic net logistic regression achieved the highest area under the receiver operating characteristic curve (0.819) and the lowest Brier score (0.170), with an accuracy of 77.4%. The constrained gradient-boosting tree achieved the highest accuracy (79.0%). Permutation analysis identified mean METs as the most consistent contributor across algorithms. These findings suggest that wearable activity measurements combined with individual characteristics provide a promising basis for predicting hazard-cue information acquisition decline during hot-weather construction work.
No takes yet. Share an insight, caveat, or question.
Yan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: