EEG has become the most commonly used signal for diagnosing and treating neurological conditions. These recording are frequently corrupted by various types of noise. Using various signal processing approaches, the artifact is typically removed as a preprocessing task. The Savitzky-Golay filtering approach was proposed in this article for removing noise from EEG signals. It is capable of lowering noise while preserving the shape and height of waveform peaks. The proposed noise reduction approach was tested using real-time EEG recordings of publicly available databases like mental arithmetic data and sleep data .The parameters like Signal to noise ratio (SNR) and mean absolute error (MAE) are used to evaluate the filtered signals. The results shown that the Savitzky-Golay (SG) filtered signals applied to the support vector machine and got good accuracy when compared to the maximum overlap discrete wavelet transform.