Epilepsy is the most prevalent neuro disorder in the world, may impair brain function or even put the patient's life in peril. Seizure control requires epilepsy prediction, which enables preventative actions to lessen harm or manage seizures. It has been shown that abnormal brain activity begins in the pre-ictal state, that occurs before a seizure begins. In this research, the pre-ictal period's temporal span was reevaluated and split into many temporal windows. Then it was suggested to use deep neural network to create a specific seizure prediction method. By making use of the strategy, the temporal dependence of the signal across several time frames throughout the pre-ictal phase is represented. Additionally, by implementing a soft threshold blurring and focusing procedure inside the neural network, seamless feature extraction is made possible. The outcomes of our approach are contrasted with those of more contemporary epilepsy prediction techniques. Our approach still has certain shortcomings when compared to the finest methods, but it also exhibits several novel ideas and benefits