<p>Epileptic seizure recognition from electroencephalogram (EEG) data plays a critical role in supporting early diagnosis and clinical decision making. However, centralized machine learning approaches raise privacy concerns when sensitive patient data is shared across institutions. This study proposes a privacy preserving epileptic seizure recognition framework using federated and explainable machine learning. Experiments were conducted using a publicly available EEG derived dataset comprising five original classes, which were reformulated into a binary classification task distinguishing seizure and non-seizure events. Data were partitioned at the patient level to prevent information leakage, with separate training and testing sets. A simulated federated learning environment was implemented with five clients, incorporating both independent and non-independent data distributions and executed over fifty communication rounds using the Federated Averaging strategy. Multiple machine learning models, including Support Vector Machine, Random Forest, Gradient Boosting, and K Nearest Neighbors, were evaluated using standardized hyperparameter configurations. Model performance was assessed using accuracy, area under the ROC curve, confusion matrices, and learning curves. Explainability was achieved through SHAP and LIME to provide insight into model decision behavior. The results demonstrate that federated learning can achieve competitive performance while preserving data privacy, and that explainable methods offer transparency in seizure classification decisions. This study serves as a proof-of-concept evaluation using an EEG derived dataset, highlighting methodological feasibility rather than clinical deployment.</p>