This chapter analyzes and evaluates various denoising techniques, including wavelet transform and moving average filter methods for removing ocular and motion artifacts from EEG signals. The performance of each technique is benchmarked in terms of signal-to-noise ratio (SNR) and normalized mean -squared error (NMSE) on available EEG databases, including Bonn and Motion-Artifact Contaminated EEG databases. Simulation results show that the wavelet transform using the SURE Shrink algorithm with the hard thresholding rule has the best performance for removing ocular artifacts in intracranial EEG. In contrast, the wavelet transform using the universal threshold shrinkage rule with hard thresholding is the preferred method for removing motion artifacts in scalp EEGs. We also present a detailed preprocessing technique for removing artifacts from intracranial electroencephalography (iEEG) data. This includes the patient-specific re-referencing method designed to eliminate common noise, improving the accuracy and reliability of the data for further analysis. This method selects the reference electrode based on the electrode placement distribution for each patient. To investigate the effect of the reference electrode location on the obtained results, the re-referencing method is compared to the common average reference method. These techniques are evaluated and compared on four patients from the European iEEG dataset. Simulation results indicate that the epilepsy prediction accuracy using an LSTM model based on the patient-specific method has a better performance with an average AUC of 0.83 compared to the same model when using the common average reference method with an average AUC of 0.81. This chapter is an essential step toward more advanced work to achieve real-time and low-cost denoising methods for energy-constrained devices.