Electroencephalographic (EEG) techniques are widely used in cognitive science, neuroscience, psychophysiology, and brain–computer Interface (BCI) research due to their non-invasive nature, portability, and high temporal resolution. However, EEG signals often suffer from contamination by non-brain electrical activities such as those from eye movements (EOG), muscles (EMG), and the heart (ECG), necessitating preprocessing to maintain a high signal-to-noise ratio (SNR) for accurate analysis. This research evaluates techniques for mitigating artifacts from oculomotor activities, particularly saccades, which are more challenging to remove than eye blinks. The primary methods for correcting these artifacts are regression-based techniques and Independent Component Analysis (ICA). Regression methods like the Gratton algorithm use EOG channels but can introduce contamination, while ICA methods such as AMICA require substantial computational resources and the careful selection of EEG channels. Moreover, recent advancements in algorithms have focused on identifying and correcting ocular artifacts in out-of-lab applications, using data from a low number of channels. Notably, EEGANet, based on Generative Adversarial Networks (GANs), stands out as a promising approach. It requires an initial training and optimization process using EOG channels. EEGANet’s performance was compared to Gratton, AMICA, SGEYESUB, REBLINCA, and MWF using publicly available datasets, with evaluation metrics including Pearson's correlation, mutual information, and frequency correlation. The results revealed that EEGANet showed a superior correction performance over frontal EEG channels, effectively identifying and correcting both horizontal and vertical eye saccade artifacts. It preserved the EEG signal's spectral characteristics across theta, alpha, and beta frequency bands, indicating minimal impact on the signal's neurophysiological content.