Classification of atrial fibrillation using multidisciplinary features and gradient boosting

Atrial fibrillation (AFib) is the most common tachyarrhythmia of the heart in adults and is associated with an increased risk for stroke and heart failure. It can be described as irregularly irregular, foci in the atrium that set up chaotic atrial circuits and irregular rapid contraction of the atrium with loss of consistent atrio-ventricular synchrony due to decremental conduction at the atrio-ventricular node. The challenge was to identify predictive features of ECG signals with variable time and spatial components. These features were extracted from 8528 single lead ECG recordings and then input to a gradient boosting classifier. The trained model could classify AFib with an F1 score of 0.83. Present in the dataset were three other rhythm classes; Normal Sinus Rhythm, Other, and Noisy. The F1 scores achieved for these classes were 0.91, 0.77, and 0.66 respectively.