Hybrydowe modele AI zoptymalizowane pod względem cech do wyjaśnialnego i uniwersalnego przewidywania epilepsji

Abstract Epilepsy a neurological condition where individuals experience recurring seizures, which developing accurate, timely prediction methods for the purpose of optimal clinical care. Machine learning has achieved some positive results in this area; however, there are limitations regarding generalized use, interpretability and quality of selected features limiting the potential for these models to be used clinically. Therefore, the focus of this report will be to evaluate and analyze detailly hybrid artificial intelligence (AI) frameworks using optimized features and XAI, specifically focusing on predicting epileptic seizures. The evaluation will include various types of data utilized in the form of: electroencephalograms (EEGs); patient information; and multimodal biomedical data. Each type of hybrid method that integrates machine learning and deep learning, and its corresponding predictive capabilities, robustness, interpretability, and capability to perform well under varied conditions in relation to multiple data environments evaluated. Feature optimization strategies utilizing meta-heuristics or dimensionality reduction methods are reviewed, also includes recent advancements in the fields of machine learning, deep learning, and hybrid architectures for EEGs; patient information; and multimodal biomedical datasets. The report focuses how explainability enhances clinical acceptance through the application of SHAP, LIME and attention mechanisms. Comparative analysis and identification of research areas requiring additional work were conducted as part of identifying areas to apply proposed generic framework for improving the efficacy of epilepsy prediction systems. Study aims to provide a structure for assessing current approaches; identify significant research challenges; and guide researchers toward development of highly reliable, easily interpretable for predicting epileptic seizures.