No abstract

No abstract

No abstract

No abstract

Atrial Fibrillation (A-FIB) is the most common heartbeat arrhythmia in the United States. It is characterized by an irregularity in electrical impulse reception, causing the atria to beat asynchronously from the ventricles. A-FIB episodes are sporadic, so symptoms may not appear during clinical visits, causing a risk of misdiagnosis which can lead to life-threatening strokes. This study seeks to utilize a range of nonlinear dynamic features to detect A-FIB through using RR-intervals, the distance between two R-peaks, in electrocardiograms (ECG). Data were retrieved from the Massachusetts Institute of Technology and Beth-Israel Hospital (MIT-BIH) A-FIB Database. 10 entropy=related estimators were used and compared using three systems of analysis. Hyperparameter tuning on all tree-based methods and some non-tree-based methods was conducted to improve the accuracy of A-FIB prediction. Tree-based methods, such as Random Forests and Cat Boosting, performed higher accuracy in A-FIB prediction than non-tree-based models. If implemented in wearable heart rate monitoring devices, such as smartwatches, the machine learning models of this study can provide a preliminary A-Fib diagnosis outside of clinical settings.