Towards an EEG-Based Approach for Detecting Falls from Height Hazards Using Construction Workers' Physiological Signals

Falls from height (FFH) are the leading cause of fatalities in construction. Traditional methods for detecting FFH hazards have great limitations due to workers' desensitized risk perception or supervisors' subjective inspection. Electroencephalogram (EEG) provides an objective metric for overcoming the limits. Early hazard detection can be facilitated by measuring individual physiological states, exhibiting atypical patterns when workers face risks. Although previous work covered hazard detection using EEG, there is a scarcity of FFH-specific emphasis. How to determine appropriate EEG features for detecting FFH hazards was still not discussed. Therefore, this paper evaluated the validity of an EEG-based approach for detecting FFH hazards by establishing five supervised machine learning models. EEG data was collected from 20 front-line construction workers. Two EEG feature selection techniques were included, and performances of five classifiers were compared. Support vector machine was found to have the best overall classification performance for FFH hazard detection when using the filter-based feature selection approach, with a high accuracy of 79.20%. Adopting the proposed approach has the potential to bring managerial benefits in proactive safety management. It sheds light on the development of an early warning system through real-time monitoring of physiological states of construction workers while working at height.