Enhanced EEGNet optimization using lightweight Deep Neural Network for detecting human stress levels from Raw EEG signals dataset

Accurate stress detection is essential in neuroscience and healthcare since human stress levels have a substantial impact on general well-being. Although EEG-based stress detection is a popular non-invasive method, real-time applications are limited by the high processing costs of classic feature extraction techniques. This work removes the requirement for intricate feature extraction by presenting an efficient EEGNet framework for real-time stress prediction utilizing raw EEG signals. The suggested approach improves accuracy and efficiency by utilizing Z-score normalization and adjusting batch size. Compared to traditional CNNs, the lightweight EEGNet architecture's depth-wise and separable convolutions ensure computational efficiency by reducing trainable parameters by a factor of ten. The proposed method outperformed current EEGNet-based techniques in stress identification by 5% to 18%. Furthermore, the improved system efficiently handles raw EEG signals, which is suitable for real- time, resource-constrained scenarios. The results shows the effectiveness of model centric optimization in deep learning and open the door to the creation of more effective stress detection techniques. This work provides a workable solution for real- time non-invasive stress monitoring, highlighting the significance of lightweight frameworks customized for particular applications.