Mobile EEG-Based Workers’ Stress Recognition by Applying Deep Neural Network

A large number of construction workers are struggling with high stress associated with their perilous job sites. Excessive occupational stress can cause serious job difficulties by negatively impacting workers’ productivity, safety, and health. The first step to decrease the adverse outcomes of this work-related stress is to measure workers’ stress and detect the factors causing stress among workers. Various self-assessment instruments (e.g., a stress assessment questionnaire) have been used to assess workers’ perceived stress. However, these methods are compromised by several drawbacks that limit their use in the field. Firstly, these methods interrupt workers ongoing tasks. Secondly, these methods are subject to a high degree of bias, which can lead to inconsistent results. The authors’ earlier work attempted to address the limitations of these subjective methods by applying different machine learning methods (e.g., Supervised Learning algorithms) to identify the pattern of workers’ brain waves that is acquired from a wearable Electroencephalography (EEG) device, while exposed to different stressors. This research thus attempts to improve the stress recognition accuracy of the previous algorithms by developing an EEG-based stress recognition framework by applying two Deep Learning Neural Networks (DNN) structures: a convolutional deep learning neural network (deep CNN) and a Fully Connected Deep Neural Network. Results of the optimum DNN configuration yielded a maximum of 86.62% accuracy using EEG signals in recognizing workers’ stress, which is at least six percent more accurate when compared with previous handcraft feature-based stress recognition methods. Detecting workers’ stress with a high accuracy in the field will lead to enhancing workers’ safety, productivity, and health by early detection and mitigation of stressors at construction sites.