Feasibility of an EEG-based dynamic suboptimal cognitive monitoring for field neuroergonomics

Abstract

Suboptimal cognitive states among construction workers significantly impact safety and productivity, with mental workload playing a key role in triggering these states. Determining if the mental workload fluctuation is leading to an error is challenging as the relationship between mental workload and suboptimal cognitive states is complex and non-linear, with traditional theories failing to map their fluctuations effectively. Recently, a two-dimensional space has been introduced to theoretically map mental workload fluctuations and suboptimal cognitive states using task engagement and arousal. However, there is currently no framework in place to continuously apply this theoretical knowledge in practical settings. To address this gap, this study investigates the feasibility of EEG-based frameworks for classifying four different cognitive states, namely comfort zone, mind wandering, effort withdrawal, and inattentional blindness, based on mental workload fluctuations. EEG signals were collected from 10 participants using a headset with dry electrodes, processed to extract relevant features, and classified using Support Vector Machine (SVM) and Artificial Neural Network (ANN) models. The ANN achieved superior performance in k-fold and leave one period out validation methods, though accuracy declined in leave one subject out validation. These findings underscore the potential of EEG-based differentiation of cognitive suboptimalities to enhance safety and productivity in construction by providing crucial information about when construction workers are most likely to make cognitive errors, which is essential for timely and appropriate interventions. Also, the low subject independent accuracy emphasizes the need to address individual differences in EEG signals for broader applicability.