Background: Electroencephalography (EEG) is non-invasive technique has the capability to detect minuscule variations in voltage that arise from the movement of ionic currents within the neurons present in the cerebral cortex. These recordings can help to diagnose brain disorders such as tumours, seizures specially epileptic seizures. But these EEG recordings are often distorted by undesired noise due to eye movements and blinking, known as ocular artifacts. Objective: Detection and removal of artifact present in EEG recordings is crucial one. These artifacts are of same signal frequencies and overlapped with pure EEG signals. During the analysis of these signals, the classification results may varies due to non-availability of artifact free signals. The proposed study is two-step process that initialize with detection and removal of ocular artifact arise due to eye blink and eye movement in UCI epileptic dataset. The deep learning based modified Gated Recurrent Unit is applied for epileptic seizure classification. Methods:The study focused on removing ocular artifacts with Independent Component Analysis - Discrete Wavelet Transforms, employing an optimized wavelet function. After successfully removing the ocular artifact, the next step involved classifying epileptic seizures using a deep learning model modified Gated Recurrent Unit (GRU). Results:The results of this study are compared to outcomes obtained from analysing the contaminated UCI epileptic EEG dataset.The findings showed that clean data produced superior results in terms of accuracy, precision, recall, and F1-score. Remarkably, the analysis demonstrates significant improvement in classification accuracy of 99.50%. Conclusion:The Modified-GRU model enhances electroencephalographybased epileptic seizure classification outcomes, demonstrating its potential for developing accurate and reliable real-world EEG-based Brain Computer Interface (BCI) and ensures the potential for continued impact in the field of medical signal processing.