EEG-based workers' stress recognition at construction sites

• Published in: Automation in construction Vol. 93; pp. 315 - 324
• Main Authors: Jebelli, Houtan, Hwang, Sungjoo, Lee, SangHyun
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01.09.2018; Elsevier BV
• Subjects: Algorithms; Brain; Brain wave patterns; Construction; Construction sites; Construction workers' stress; Electroencephalogram (EEG); Electroencephalography; Machine learning; Occupational health; Occupational stress; Recognition; Signal quality; Stresses; Supervised learning; Support vector machines; Wearable biosensors in construction sites; Workers' health, safety, and productivity; Construction workers' stress; Brain wave patterns; Electroencephalogram (EEG); Wearable biosensors in construction sites; Supervised learning; Workers' health, safety, and productivity
• ISSN: 0926-5805; 1872-7891
• DOI: 10.1016/j.autcon.2018.05.027

Taking into account that many construction workers suffer from excessive stress that adversely impacts their safety and health, early recognition of stress is an essential step toward stress management. In this regard, an electroencephalogram (EEG) has been widely applied to assess individuals' stress by analyzing brain waves in the clinical domains. With recent advancements in wearable EEG devices, EEG's ability can be extended to field workers, particularly by non-invasively assessing construction workers' stress. This study proposes a procedure to automatically recognize workers' stress in construction sites using EEG signals. Specifically, the authors collected construction field workers' EEG signals and preprocessed them to capture high-quality signals. Workers' salivary cortisol, a stress hormone, was also collected to label low or high-stress levels when they work at sites. Time and frequency domain features from EEG signals were calculated using fixed and sliding windowing approaches. Finally, the authors applied several supervised learning algorithms to recognize workers' stress while they are working at sites. The results showed that the fixed windowing approach and the Gaussian Support Vector Machine (SVM) yielded the highest classification accuracy of 80.32%, which is very promising given the similar accuracy of stress recognition in clinical domains where extricate and wired EEG devices were used and the subjects engage in minimal body movement. The results demonstrate that the proposed field stress recognition procedure can be used for the early detection of workers' stress, which can contribute to improving workers' safety, health, wellbeing, and productivity. •A procedure to recognize workers' stress using a wearable EEG device is developed.•The procedure extracts a broad range of EEG features in time and frequency domains.•The procedure applies supervised learning algorithms to detect stress.•The performance of the procedure was examined on the EEG signals collected from the real construction sites.•Proposed stress recognition procedure is competitive with other stress recognition algorithms in the clinical domain.