EEG signal cleaning for brain wave monitoring of construction workers using prediction and clustering models

The use of electroencephalogram (EEG) signals is increasingly recognized as an effective method for assessing the workload and risk awareness of construction workers. However, due to sensor malfunctions and environmental interference, EEG signals often contain significant anomalies that hinder accurate cognitive analysis. This study proposes a robust EEG signal cleaning framework to automatically detect and reconstruct such abnormal data. The framework combines a fusion attention mechanism with a bidirectional gated recurrent unit (ATT-BiGRU) model for EEG signal prediction and employs the density-based spatial clustering of applications with noise (DBSCAN) algorithm for anomaly detection. The predicted data are then used to reconstruct missing or corrupted signal segments. A spatial perception experiment simulating AR-assisted indoor navigation and measurement tasks is conducted to validate the framework. Abnormal EEG data are categorized into four types: outliers, drift, missing, and minor signals. The ATT-BiGRU model achieves a prediction RMSE below 0.05, while the DBSCAN algorithm demonstrates over 96 % detection accuracy across multiple EEG channels. Additional validation using data from 20 participants shows consistent accuracy above 94 %. These results highlight both the technical efficacy and scalability of the proposed framework, as well as its potential for supporting cognitive monitoring in complex-built environments. By enhancing EEG signal reliability, the framework enables more effective tools for architectural cognition assessment, informs human-centered design strategies, and lays the foundation for adaptive environments that respond to users’ real-time neurophysiological states.