Abstract
In the area of smart healthcare services in IoT-connected smart cities, the integration of cutting-edge CNN models such as DenseNet201, ResNet152V2, and MobileNetV2 with innovative IoT technology for automated seizure detection and monitoring through EEG data analysis. Leveraging a three-layer architecture encompassing device, edge server, and global cloud layers, this study has introduced a novel method to monitor patients' health conditions. The implementation of a cropping training strategy enhances deep learning model efficiency in scenarios with limited data. Notably, the classification accuracy of 2 level CNN models, particularly DenseNet201, reaches remarkable heights, with seizure detection rates ranging from 99.26% to 99.51%, and non-seizure accuracy spanning 99.25% to 99.46%. These models excel in precision for seizure identification, with values between 99.33% and 99.53%, and they exhibit robust recall, particularly 2 level CNN DenseNet201 with 99.24%. The use of Matthews Correlation Coefficient (MCC) further affirms their precision and robustness, with values ranging from approximately 98.53% to 99.19%. This approach, combining advanced CNN models with IoT technology, holds great promise for efficient seizure detection and monitoring, while the assessment of energy consumption and task processing times underlines the importance of model selection and edge server configurations in optimizing system performance.