The electroencephalogram (EEG) has been commonly used to identify epileptic seizures, but identification of seizures from EEG remains a challenging task that requires qualified neurophysiologists. It is important to detect seizures in real time, which can be achieved in an internet of things (IoT)-based cloud platform to alert patients of impending seizures. Therefore, in this study, we propose a new model to recognize seizure states (e.g., ictal, preictal, interictal) from EEG in the IoT framework to monitor patients remotely. The proposed model uses an efficient hybrid genetic whale optimization algorithm (GWOA) based on naïve Bayes (NB-GWOA) for feature selection, and an adaptive extreme learning machine (ELM) based on a differential evolutionary (DE) algorithm (DEELM) for classification. In the NB-GWOA method, the genetic algorithm serves to enhance the exploitation of the whale optimization algorithm in the search of the optimal solutions, while the naïve Bayes method is used to determine a fitness function to assess every agent in the search space. GWOA has strong robustness and is capable of finding the best solutions in less than five iterations, so it is suitable for selecting discriminative features from a huge number of neurofeatures obtained from EEG. Further, the classification model is constructed based on ELM, which uses the DE algorithm for a fast and efficient learning solution. Results show that the proposed NB-GWOA-DEELM model can avoid over- and under-fitting and can provide better and more accurate performance in classifying seizure states from EEG than its competitors.

脑电图 (EEG) 已常用于识别癫痫发作，但从 EEG 中识别癫痫发作仍然是一项具有挑战性的任务，需要合格的神经生理学家。实时检测癫痫发作非常重要，这可以在基于物联网 (IoT) 的云平台中实现，以提醒患者即将癫痫发作。因此，在本研究中，我们提出了一种新模型，可以在 IoT 框架中从 EEG 中识别癫痫发作状态（例如，发作期、发作前期、发作间期）以远程监测患者。所提出的模型使用基于朴素贝叶斯 (NB-GWOA) 的高效混合遗传鲸鱼优化算法 (GWOA) 进行特征选择，并使用基于差分进化 (DE) 算法 (DEELM) 的自适应极限学习机 (ELM) 进行分类. 在NB-GWOA方法中，遗传算法用于增强鲸鱼优化算法在搜索最佳解决方案中的利用，而朴素贝叶斯方法用于确定适应度函数以评估搜索空间中的每个代理。GWOA 具有很强的鲁棒性，能够在不到 5 次迭代中找到最佳解，因此适用于从 EEG 获得的大量神经特征中选择判别性特征。此外，分类模型是基于 ELM 构建的，它使用 DE 算法进行快速有效的学习解决方案。结果表明，所提出的 NB-GWOA-DEELM 模型可以避免过拟合和欠拟合，并且在从 EEG 分类癫痫发作状态方面比其竞争对手提供更好、更准确的性能。