This study presents an EEG-based authentication framework that adopts the previously Compact Data Learning (CDL) methodology for data-efficient optimization. EEG signals, characterized by high dimensionality and inter-individual variability, were preprocessed and analyzed using feature extraction and reduction techniques to enhance discriminative capability while reducing computational complexity. The adopted CDL framework integrates feature and sample reduction to construct compact datasets that preserve essential information, significantly improving training efficiency. Four machine learning models—XGBoost, SVM, LSTM, and CNN—were evaluated on the BS-HMS dataset. Experimental results demonstrate that the combined CDL framework reduced data volume to only 8.1% of the original dataset while maintaining high recognition accuracy. The CNN model achieved the best overall performance, reaching an accuracy of 87.50% with a 73.5% reduction in training time. These findings validate the CDL framework as an effective approach to optimizing EEG-based biometric systems, enhancing their scalability and applicability in computationally constrained environments.
Zhao et al. (Sun,) studied this question.