The construction industry is among the most stressful of occupations. Traditional techniques to evaluate worker stress (e.g., self-assessment and observational checklists) may not be effective in the field because of their potential to interrupt workers' ongoing tasks. Additionally, these methods are subject to biases. Research in neuroscience confirms that the human brain responds to various stressors, so studying patterns of brainwave activity can lead to strong assessments of subjects' stress. The authors' earlier research has monitored worker stress using a wearable electroencephalography (EEG) headset by applying supervised learning algorithms in a binary stress level (i.e., low and high). Despite the success of earlier work identifying excessive stress, there is a gap in knowledge in assessing medium-level stresses. It has been proven that intermittent exposure to medium-level stress decreases not only performance but also concentration and focus. This research attempts to identify multiple levels of worker stress using signals recorded from a wearable EEG headset by applying two supervised learning algorithms, multi-class support vector machines (SVM) and fully connected neural network (FCNN). A stress-related hormone, cortisol, was used as a baseline to label subjects' stress levels. In classifying three levels of stress, the FCNN yielded a prediction accuracy of 79.26%, which is competitive with previous EEG-based stress recognition methods in a binary setting. This research should help in identifying multiple levels of stress at construction sites and aid early detection and mitigation of high stress in the field.