Epileptic seizures are sudden neurological events that might cause dire consequences to the well-being and health of a patient. To achieve timely interventions, risk mitigation, and better clinical outcomes, sufficient seizure identification and predictive opportunities must be evident. This research aims to develop an explainable deep learning model that can not only classify EEG data into seizures and non-seizures. One of the most important contributions of this work is introduction of a new multimodal deep learning (DL)structure, which depends on the principles of the electroencephalogram (EEG) signal prediction of seizures. Temporal Fusion Transformer (TFT) - a time-series forecasting model, is the most advanced in the framework. The TFT architecture is especially helpful in this task because it can capture time dependencies over a long range, missing data, and multi-horizon prediction. Its attention-based mechanism also allows it to be interpretable, to determine important time windows and biomarkers preceding a seizure, and to give useful information about the seizure onset dynamics. To make the model transparent and clinically relevant, the framework is based on Shapley Additive Explanations (SHAP) as an established explainable AI (XAI) method. The model decodes the decisions and visualizes them using SHAP, which contributes to a better chance of healthcare professionals trusting and understanding the predictions. The proposed system provides a highly interpretable, multimodal, and efficient solution to the problem of real-time seizure monitoring. It supports patient-centered, clinically deployable seizure prediction models that have the potential to play an essential role in personalized healthcare and enhance the quality of life(QoL) of epilepsy patients.