A comparative and critical study of EEGNet for fNIRS-driven cognitive load classification

This study evaluates EEGNet for classifying cognitive load from fNIRS signals, focusing on various temporal segmentation strategies, window lengths, feature extraction methods, and learning rate configurations. Key findings indicate that overlapping segmentation with smaller fixed learning rates (0.01-0.001) yields the highest accuracy in random-split evaluations, while non-overlapping segmentation outperforms in subject-independent evaluations, achieving a maximum accuracy of 56.11% with PCA features and a 20-second window. These results underscore the importance of segmentation strategy and learning rate in enhancing model generalization, which is critical for developing reliable cognitive load classification systems in real-time applications.