Seizure Detection Via Time Series Classification Using Modified Metaheuristic Optimized Recurrent Networks

Epilepsy, colloquially termed seizure disorder, constitutes a neurological condi-tion characterized by unpredictable and sudden episodes of heightened electrical activity in the brain. The inherent risks associated with seizures, particularly in their nascent stages, stem from their challenging predictability, often leading to individuals endangering themselves through falls or lapses into unconsciousness in precarious settings. Additionally, the potential for neurological damage further compounds the complexity of managing this condition. This research endeavors to explore the viability of Recurrent Neural Networks (RNNs) for the early detection of seizures using electroencephalogram (EEG) recordings derived from real-world scenarios. Acknowledging the pivotal role of network topology and training parameters in achieving optimal performance, this study employs meta-heuristic algorithms to fine-tune performance through judicious hyperparameter selection. Notably, a modified iteration of a recently introduced algorithm is introduced herein. The application of RNNs to seizure detection yields promising outcomes, with constructed models achieving an accuracy surpassing 99%. The results obtained underscore the potential of employing metaheuristic-optimized RNNs in seizure detection as a means of substantially enhancing the quality of life for individuals afflicted by epilepsy. The demonstrated accuracy of these models suggests a robust and reliable methodology, paving the way for future advancements in the realm of neurotechnological interventions. This investigation contributes valuable insights into the intersection of machine learning, neuroscience, and healthcare, offering a foundation for further research and technological developments aimed at ameliorating the challenges posed by epilepsy.