Epileptic Seizure Classification Based on Enhanced State Refinement Gated Recurrent Unit with Temporal Activation Regularization

The neural activity in the brain is detected through Electroencephalography (EEG) which enables the analysis and classification of epileptic disorder.The epileptic classification is challenging due to the presence of noise and artifacts in the EEG signal which increases the False Positive Rate (FPR) and minimizes the classification performance.Therefore, this research proposes an Enhanced State Refinement Gated Recurrent Unit with Temporal Activation Regularization (ESRGRU-TAR) for epileptic seizure classification.The ESRGRU optimizes the gating mechanism to enhance the capability to capture long-term dependencies in the data.Particularly, refinements are capable of quality of interactive models that reveal interactions among sample points thereby enhancing interpretability.The message-passing mechanism is developed to highlight useful feature representations between sample points.TAR is beneficial for controlling overfitting because it moderates the model's activation by adding temporal consistency to the learning process.The BONN and CHB-MIT datasets are used to estimate the proposed ESRGRU-TAR performance based on the classifier.The ESRGRU-TAR achieves better accuracy of 99.91% and 99.89% for BONN and CHB-MIT datasets which is better than existing techniques such as Bidirectional Long Short-Term Memory (Bi-LSTM).