<![CDATA[Cardiovascular disease (CVD) is still the leading cause of death worldwide, and arrhythmia is one of its most serious forms because it can trigger sudden cardiac arrest. Given the life-threatening nature of arrhythmias, reliable automated methods for arrhythmia detection are increasingly important in both clinical and remote monitoring settings. While the electrocardiogram (ECG) is the standard tool for arrhythmia detection, its accuracy is often reduced by noise and waveform distortion, which may lead to misclassification. To address this challenge, this study proposes an arrhythmia classification framework that integrates wavelet-based feature extraction with filter bank enhancement. ECG signals from the MIT-BIH Arrhythmia Database were preprocessed and segmented from two leads (MLII and V1), followed by wavelet decomposition using Daubechies (db6), Symlet (sym7), and Biorthogonal (bior4.4) families. Three complementary feature enhancement schemes, Discrete Cosine Transform (DCT), Complex Discrete Wavelet Transform (CDWT), and Orthogonal filter bank, were applied prior to classification with Support Vector Machine (SVM) and Random Forest (RF). The experimental results further highlight that the selection of wavelet, filter bank, and classifier combinations significantly influences arrhythmia detection performance. In particular, the pairing of the bior4.4 wavelet with the orthogonal filter bank and RF classifier achieved the highest accuracy of 94.76%, outperforming other setups, including CDWT-based schemes. This outcome suggests that the linear phase property of bior4.4 yields a more stable feature representation that aligns well with the ensemble mechanism of RF. These insights reinforce the importance of considering both the mathematical properties of wavelets and classifier design when developing ECG-based diagnostic systems. Future work will extend this approach to deep learning models and larger datasets to strengthen its clinical applicability.]]>
Copyrights © 2026
