ECG arrhythmia classification using cyclic transformations and regression-based features

The clinical importance of detecting cardiac diseases has led to ongoing efforts to improve classification algorithms used in electrocardiography (ECG) analysis. Because ECG signals are patterned, non-stationary, and complex, representing them in cyclic form and extracting features designed for this representation may improve classification performance compared to conventional methods. In this study, the ECG signals were transformed into cyclic form and this transformation was validated using statistical tests. The transformed signals were then classified using features specifically developed for cyclic data. At the same time, classification was also performed on the original raw, untransformed signals to enable comparison. In addition, the study introduced a regression-based feature set as a second novel contribution. This feature set was designed to work with both cyclic and non-cyclic signal representations, making it suitable for a broader range of analysis conditions. The classification results were evaluated across four transformation cases and three sets of characteristics, providing a comprehensive analysis of ECG signal classification. An important strength of this study is that the labels correspond to specific types of arrhythmia. This allows the evaluation to remain clinically grounded and supports a more meaningful assessment of how well different classes of ECG signal can be distinguished. In general, the proposed framework provides clinically relevant findings and may advance the detection of ECG-based arrhythmias.