A novel epileptic seizure prediction approach for medically refractory epilepsy patients using BrainComputer Interface (BCI), which is cloud based is developed in this work. Using this approach, the strategically implanted electrodes are influenced to effectively reselect the site of epileptogenicity, however owing to the immediacy of electrographic ictal behaviour, obtaining real time solution is a challenge. The Electrocorticographic (ECoG) and Electroencephalographic (EEG) signals are highly non-linear in nature resulting in wide ictal and normal pattern deviations. It is unfeasible to apply manually derived features for detecting seizure, since even a limited set of electrodes generate vast amount of data that takes longer time to process. Hence, by harnessing the capabilities of both deep learning and cloud computing, a seizure detection approach using BCI as Internet of Things (IoT) is designed and implemented in this paper. The process of feature extraction in addition to classification is carried out using Convolution Neural Network (CNN). The weights and biases of the CNN are optimized using Whale Optimization Algorithm (WOA). Using IoT, the storage, automatic computing and real time processing of the proposed approach is implemented. The electroencephalogram (EEG) has been widely used to detect epileptic seizures; however, it is still difficult to detect seizures from the EEG and necessitates the expertise of neurophysiologists. Real-time seizure detection is crucial for warning patients of imminent seizures, and it is possible using an Internet of Things (IoT)-based cloud platform. The electroencephalogram (EEG) has been widely used to detect epileptic seizures; however, it is still difficult to detect seizures from the EEG and necessitates the expertise of neurophysiologists. Real-time seizure detection is crucial for warning patients of imminent seizures, and it is possible using an Internet of Things (IoT)-based cloud platform. Thus, it is appropriate for choosing discriminative characteristics from a vast array of neurofeatures derived from EEG. The classification model is also built using the ELM framework, which leverages the DE algorithm for quick and effective learning. Findings indicate that the proposed NB-GWOA-DEELM model outperforms its rivals in categorising seizure states from EEG by avoiding over- and under-fitting and by performing more accurately.