Spatio Temporal Attentional EEGNet: An Enhanced Deep Learning Model for Cognitive Workload Detection

Abstract

Deep learning has emerged as a powerful tool for extracting meaningful patterns from electroencephalography (EEG) signals, particularly for mental workload recognition in cognitive neuroscience. This study proposes STA-EEGNet (Spatio Temporal Attentional EEGNet), an enhanced variant of the EEGNet architecture that integrates spatio temporal attention mechanisms to improve EEG signal decoding. The model is trained and evaluated on the STEW: Simultaneous Task EEG Workload Dataset, demonstrating superior classification accuracy compared to existing approaches. Heatmap analyses of key convolutional filters in TSA-EEGNet network during testing reveal distinct spatiotemporal activation patterns associated with varying workload levels. High workload is characterized by increased activation in discriminative filters and reduced or inverted responses in suppressive filters, primarily over frontocentral and parietal regions. These findings highlight TSA-EEGNets ability not only to achieve state-of-the-art performance but also to provide interpretable insights into the neural dynamics underlying cognitive workload.