A Lightweight Detection Method for Atrial Fibrillation Using Clustering Feature Extraction on Wearable Devices

Atrial fibrillation (AF) is a common arrhythmia. The accuracy and efficiency of detection for wearable ECG devices are limited by noise interference and poor data quality in realtime monitoring scenarios. This study proposes a lightweight AF detection method utilizing clustering feature extraction on wearable ECG devices. The method employs clustering techniques to obtain high-quality heartbeat waveforms and RR intervals, subsequently extracting crucial temporal and interaction features. Unlike previous methods that directly extracted features from raw heartbeat waveforms and RR intervals, this approach effectively processes noisy data from wearable devices. An eXtreme Gradient Boosting (XGBoost) classifier is trained on these features to categorize ECG recordings as normal sinus rhythm, AF, other arrhythmias, or excessive noise. The lightwright XGBoost algorithm, combined with clustering feature extraction, significantly reduces resource consumption and memory usage while maintaining high computing performance. This method is applicable to ECG applications, particularly wearable medical devices, for classifying AF and other arrhythmias. The model achieved an Fl score of 0.847±0.008 on the single-lead ECG dataset, outperforming traditional methods. The proposed algorithm is integrated into an AF detection System, providing a reliable and efficient platform for health management by accurately Processing wearable device data.