A Deployment-Oriented Human-in-the-Loop Framework for Reliable ECG Rhythm Classification Using Confidence-Guided Decision Routing
Description
This dataset contains extracted electrocardiogram (ECG) feature descriptors used for supervised rhythm classification and deployment-oriented Human-in-the-Loop (HITL) evaluation. The repository consists of two CSV files (labelled1.csv and labelled2.csv): • labelled1.csv (3,592 samples): Contains labelled ECG feature records covering clinically relevant cardiac rhythm categories, including Normal Sinus Rhythm, Atrial Fibrillation (AFib), and Sinus Tachycardia/Bradycardia. This dataset is intended for model training, cross-validation, and feature optimization. • labelled2.csv (864 samples): Contains independent ECG feature records used for deployment-oriented inference, confidence-guided decision routing (ALLOW/REVIEW), and expert verification testing. Each sample is represented by 34 handcrafted ECG descriptors capturing statistical, temporal, spectral (frequency-domain), and waveform morphology characteristics. All records are fully anonymized and contain no personally identifiable patient information.