Feasibility Study on Microstate Analysis to Enhance Generalizability in EEG Research of Construction Workers’ Cognitive States

The safety and productivity of construction projects are closely tied to workers’ cognitive states. Traditional methods for assessing these states often suffer from self-report bias and lack dynamic tracking. Electroencephalogram (EEG) provides an objective, real-time alternative, but its utility is limited by poor generalizability due to inter-individual variability. EEG microstate analysis, which converts EEG signals into cross-population brain network patterns, may buffer these differences and improve model performance. This study investigates the feasibility of using EEG microstates to enhance the generalizability of cognitive state classification among construction workers. EEG data labeled as desirable or undesirable cognitive states were collected during cognitively demanding tasks. Deep learning models were trained with and without EEG microstates using leave-one-subject-out validation. Results show that integrating microstates consistently improved classification accuracy, even with identical data volumes. These findings highlight the value of microstates in capturing individual variability and advancing EEG-based cognitive monitoring in construction settings.