HHEEG: Healthy Hybrid EEG-Based Mental Workload Assessment for Healthcare Using Integrated Spatial, Temporal Spectral Deep Learning

Abstract Recognizing and understanding human mental workload while performing a task is extremely important for enhancing cognitive state monitoring, optimizing task design, and facilitating decision-making in complex environments. Electroencephalography (EEG) provides a non-invasive technique for monitoring brain activity patterns, albeit with challenges stemming from considerable noise and dimensional complexity that impact classification performance. In recent years, deep learning methodologies have demonstrated exceptional capability in extracting sophisticated spatio-temporal patterns from EEG signals during workload classification tasks. Expanding on these advancements, this paper proposes a comprehensive deep learning framework that combines one-dimensional Convolutional Neural Networks (1D-CNN), Bidirectional Long Short-Term Memory (BiLSTM) networks, and attention mechanisms to simultaneously capture spatial, temporal, and contextual features directly from EEG signals’ raw data. Concurrently, a parallel spectral-processing path extracts frequency-domain features via Welch’s method to further boost performance in terms of spectral representation. The model was trained and tested on the publicly available STEW (Simultaneous Task EEG Workload) dataset, with a classification accuracy of 94.13% to separate low and high workload conditions. The fusion of multi-domain features and the model’s attention-driven focus on salient time windows is responsible for its better performance compared to single-stream neural models and classic machine learning. This work highlights the capability of hybrid deep architectures in real-time cognitive state monitoring and provides a basis for EEG-based adaptive systems.