AF-Intellect: Intelligent Classification System for Atrial Fibrillation

The increasing prevalence of cardiac arrhythmias—particularly Atrial Fibrillation (AF), which significantly raises the risk of stroke and other cardiovascular complications—highlights the urgent need for efficient, real-time detection systems suitable for wearable healthcare devices. This paper presents a lightweight, low-latency AF detection approach designed specifically for resource-constrained, wearable environments. To achieve this, three efficient machine learning algorithms—K-Nearest Neighbors (KNN), XGBoost, and Random Forest (RF)—were utilized for their rapid inference and low computational overhead. The proposed system introduces an R-peak-based pipeline for data segmentation and feature extraction. This approach enables precise segmentation of cardiac cycles, compared to traditional methods. Comprehensive signal preprocessing, including bandpass filtering, normalization, and resampling, was applied to enhance signal quality and ensure consistent input for classification. Experimental results showed that the proposed system is highly effective. Compared to other feature extraction techniques and deep learning models, it achieved a higher classification accuracy of 98.9 % while keeping latency lower, making it a promising solution for real-time, wearable AF detection.