Explainable Fuzzy Deep Learning for Prediction of Epileptic Seizures Using EEG

Addressing the challenge posed by the unpredictable and recurrent nature of epileptic seizures, which stand among the most significant neurological conditions, remains imperative, especially within settings inundated with high patient flow. The prompt identification of these seizures is paramount for effective patient care. Unfortunately, existing epilepsy seizure detection systems encounter limitations in availability and interpretability, thereby constraining their reliability and widespread application. Presently, neurophysiologists heavily rely on visually interpreting electroencephalogram (EEG) recordings displayed on screens to identify seizures. This article introduces an innovative method dedicated to detecting epileptic seizures within EEG signals, leveraging a specifically tailored fuzzy deep learning (FDL) architecture. The proposed methodology encompasses crucial stages of preprocessing and feature extraction, augmented by the utilization of explainable artificial intelligence models, such as local interpretable model-agnostic explanations (LIME) and Shapley additive explanation (SHAP) for enhancing model interpretability. The developed FDL model demonstrates promising results, achieving a noteworthy accuracy of 92.57%, precision of 0.96 for "normal" and 0.89 for "abnormal," recall of 0.91 for "normal" and 0.94 for "abnormal," and F1-score of 0.93 for "normal" and 0.91 for "abnormal," affirming its robustness in classification tasks. In addition, to validate the effectiveness of the proposed FDL, comparisons are performed with long-short term memory networks and 1-D convolutional neural network model models. The integration of LIME and SHAP significantly enhances the interpretability of the model, providing valuable insights into influential features. This comprehensive framework adeptly balances accuracy and interpretability, thereby making a substantial stride in advancing EEG-based diagnostic tools.