Wavelet transformation approaches for prediction of atrial fibrillation

Abstract

Prediction of atrial fibrillation (AF) is a major issue in medicine. This is due to the fact that AF is often asymptomatic. In this work, we present approaches based on wavelet decomposition to find features in the signal that can predict this disease. Our model consists of four parts: pre-processing, feature extraction, feature selection, and classification for prediction. The presented work shows a good predictive performance (94% accuracy) before 5 min of AF onset and a prediction accuracy of 85.5%, 110 min before AF onset. Our code will be available for researchers upon request.