ConPPMF: Construction Datasets for Privacy-Preserving Mental Fatigue

Construction workers’ unsafe behavior is the leading cause of on-site accidents, greatly influenced by mental fatigue. Existing available datasets for mental fatigue are scarce in construction. Moreover, collecting and using private data, such as facial images, bio-signals, and speech, raises significant privacy concerns, leading to difficulty in data collection during model implementation. To tackle these challenges, this research aims to introduce a replicable multi-modal dataset named ConPPMF to monitor mental fatigue on construction sites, focusing on worker comfort and privacy. The dataset comprises physiological data from smartwatches and time-synchronized visual data from cameras. Fatigue status is labeled through self-reported assessments and on-the-job performance. The dataset includes 92 videos from 17 male workers in various trades, such as crane operators, truck drivers, excavators, and pavers, collected at three construction sites during actual work hours. The dataset has been rigorously evaluated under eight quality metrics: data integrity, interpretation, lineage, accessibility, accuracy, reliability, relevance, and privacy. Besides, ConPPMF achieved an F1-score of 0.84 and a recall of 0.95 in predictive models, outperforming YawDD and DROZY. These findings highlight the feasibility and ethical considerations of the ConPPMF dataset for real-world mental fatigue monitoring in construction. This research contributes to developing robust, privacy-preserving tools for early fatigue detection, ultimately enhancing construction safety and reducing on-site accidents.