Comparison of moving average and Savitzky-Golay filters for noise removal in electrocardiograms

This paper presents the results of independently applying two filters to eliminate noise from an ECG signal: Moving Average Filter (MAF) and Savitzky-Golay Filter (S-G). These filters were applied to 45 ECGs from the Physionet database, all diagnosed with sinus bradycardia. A Python script was developed for automatic application of the filters using the following parameters: MAF was applied using three window sizes (7, 14, and 21), and S-G with a window size of 21 and polynomial orders of 7, 9, and 11. Results indicate that MAF performs best with a window size of 7. However, the S-G filter (21,11) significantly outperforms MAF in noise removal. Statistical parameters used were MSE (MAF = 0.0435; S-G = 0.0062), SNR (MAF = 16; S-G = 25), and Correlation Coefficient (MAF = 0.988; S-G = 0.9944); these values indicate that the S-G filter achieves better noise removal and minimal signal distortion, making it the best option for such processes. It is important to note that six filter configurations were compared, with S-G (21,11) being the best. A test hypothesis points out that there are meaningful differences between SG (21,11) and MAF (7).