Comparative Analysis of EEGNet, EEG Conformer, and Spiking Neural Networks for Mental Workload Classification

Mental workload assessment is important for intelligent systems designed to monitor human cognitive state. This study presents a comparative analysis of EEGNet, EEG Conformer, and Spiking Neural Networks (SNNs) for binary mental workload classification using the Simultaneous Task EEG Workload (STEW) dataset. EEG signals were processed using fixed-length segmentation, artifact detection, subject-wise data partitioning, and training-set-based Z-score normalization. All three architectures were evaluated under a consistent preprocessing and evaluation framework using accuracy, precision, recall, F1-score, and ROC-AUC. EEGNet achieved the highest performance with 94.14% accuracy, followed by EEG Conformer at 62.61% and SNN at 53.60%. The findings indicate that the compact EEGNet architecture provided stronger generalization under the experimental conditions of this study.