RR Interval-based Atrial Fibrillation Detection using Traditional and Ensemble Machine Learning Algorithms.

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Atrial fibrillation (AF) is a serious condition linked to stroke and heart failure.
• Early detection of AF is vital to reduce mortality and morbidity.
• Existing methods require comparison with advanced machine learning techniques.

Purpose of the Study:

• To propose and compare traditional machine learning (TML) and ensemble machine learning (EML) algorithms for early AF detection.
• To evaluate the performance of various TML and EML classifiers using ECG data.
• To identify the most effective method for accurate AF classification.

Main Methods:

• Extraction of RR interval features from electrocardiogram (ECG) data.
• Classification of heart rhythms into normal, AF, and other categories using TML (e.g., SVM, KNN) and EML (e.g., Random Forest, Rotation Forest) algorithms.
• Evaluation using the PhysioNet challenge 2017 dataset with tenfold cross-validation.

Main Results:

• The Random Forest (RF) classifier achieved a high classification accuracy of 99.10%.
• The Area Under the Curve (AUC) for the RF classifier was 0.998, indicating excellent performance.
• EML methods, particularly RF, demonstrated superior performance compared to TML techniques.

Conclusions:

• Ensemble learning, specifically the Random Forest algorithm, offers a highly effective approach for accurate AF detection.
• The proposed methodology provides a robust tool for healthcare management systems, including pacemakers and defibrillators.
• This study validates the superiority of EML over TML for reliable AF identification from ECG signals.