The proposed SA-NSGA-II multi-objective optimization framework achieved up to 99.83% seizure classification accuracy using only 8 channels and 12 frequency bands, outperforming standard NSGA-II.
Does joint optimization of EEG channels and frequency bands using SA-NSGA-II improve resource-efficient seizure classification?
The proposed SA-NSGA-II framework effectively balances seizure classification performance with EEG acquisition and computational costs, facilitating resource-efficient real-time EEG monitoring.
Objective With the rapid development of wearable electroencephalogram (EEG) devices, the epileptic seizure classification system is required to deliver reliable performance under real-time and resource-constrained conditions. To this end, this study aims to reduce EEG signal acquisition and processing costs while maintaining seizure classification performance in order to facilitate the clinical deployment of intelligent EEG analysis systems. Methods We jointly optimize the number of EEG channels and frequency bands, with the goals of maximizing classification performance while minimizing signal acquisition and computational costs. The proposed optimization problem is solved by the structure-aware non-dominated sorting genetic algorithm II (SA-NSGA-II). The random forest method is employed as the classifier for seizure classification. Experiments are conducted using the public CHB-MIT scalp EEG database. Results Among the optimal channel-and-frequency-band configurations, channels P3-O1, P4-O2, and CZ-PZ are selected with high frequencies, indicating their high relevance for seizure classification. Furthermore, the gamma and alpha bands account for the largest two selection proportions, which suggests their key roles in optimal configurations. In addition, the proposed SA-NSGA-II method demonstrates effective performance in the EEG channel-and-frequency-band selection. Conclusion The proposed framework effectively balances classification performance with EEG acquisition and computational costs. By jointly selecting channels and frequency bands, our method provides an easy-to-implement solution for resource-efficient seizure classification in real-time EEG monitoring.
Chen et al. (2026) studied Epileptic seizure (n=23). Structure-aware non-dominated sorting genetic algorithm II (SA-NSGA-II) vs. Standard NSGA-II was evaluated on Seizure classification accuracy. The proposed SA-NSGA-II multi-objective optimization framework achieved up to 99.83% seizure classification accuracy using only 8 channels and 12 frequency bands, outperforming standard NSGA-II.