Epilepsy is a neurological disorder marked by abnormal brain activity. Normally, brain cells (neurons) communicate with each other through electrical signals. In epilepsy, these signals become disrupted, leading to excessive and synchronized firing, resulting in seizures. But, if it can be predicted early, then it may be prevented or made to progress slowly. For prediction, machine learning has played a great role nowadays. In this study, we have used models like Random Forest, Decision Tree, Extra-trees classifier, Gradient Boost (XGBoost) has given better result for prediction. In this paper we have taken total 500 individuals EEG data with each has 4097 data points. The EEG scans provided detailed images of brain structures, allowing for precise analysis of epileptic activity using EEG data. The prediction of seizures has proven to be a formidable challenge, with existing studies failing to yield consistently accurate results. Our methodology involved a 60/40 train-test split and again 20/20 for test-validation split ensuring the model was trained on the majority of the data. By utilizing the XGBoost, we achieved an accuracy of 98%, demonstrating its effectiveness in handling complex spatial data inherent in EEG scans. Besides, we have presented roc-auc curve, f1-score, recall, support, precision and confusion matrix to show the execution of the model. This localization capability represents a significant advancement in epilepsy research, enabling more targeted treatment strategies. The researches already performed on prediction of Epilepsy disease has given poor performance of 95% classification accuracy in comparison to our model.