Machine learning methods for detecting atrial fibrillation using electrocardiogram and photoplethysmogram
Author(s)
Wong, Kai Yuen
Soni, Aakash
Shukla, Pancham
Zhou, Keming
Type
Journal Article
Abstract
AtrialFibrillation (AFib) is a prevalent heart disorder characterised by an irregular and often
occurring rapid heart rhythm. It poses significant health risks, especially among the aging population.
Traditional manual diagnosis is time consuming and resource intensive. Fortunately, modern wearable
technology makes it possible to record and process the cardiac data outside clinical settings and facilitate
automatic diagnosis through cloud-based machine learning (ML) models. This paper proposes the use
of novel features such as peak intervals, instantaneous frequency and spectral entropy, extracted from
electrocardiogram (ECG) and photoplethysmogram (PPG) data, to detect AFib using machine learning
(ML) models. We utilise preprocessing techniques to remove artifacts from the raw data. We then use the
preprocessed data to extract the key features. These features and the raw data together are then used to train
and evaluate various ML models, including support vector machines (SVM), convolutional neural networks
(CNN), and long short-term memory (LSTM). Metrics such as accuracy, F1 score, and receiver operating
characteristic (ROC) are employed to evaluate the performance of our approach for AFib detection. Through rigorous experimentation, we identify the combination of features and ML models that are promising in quick and reliable AFib detection. By comparing related works on PPG, we also investigate whether PPG is viable in aiding or replacing ECG as a data source for AFib detection.
occurring rapid heart rhythm. It poses significant health risks, especially among the aging population.
Traditional manual diagnosis is time consuming and resource intensive. Fortunately, modern wearable
technology makes it possible to record and process the cardiac data outside clinical settings and facilitate
automatic diagnosis through cloud-based machine learning (ML) models. This paper proposes the use
of novel features such as peak intervals, instantaneous frequency and spectral entropy, extracted from
electrocardiogram (ECG) and photoplethysmogram (PPG) data, to detect AFib using machine learning
(ML) models. We utilise preprocessing techniques to remove artifacts from the raw data. We then use the
preprocessed data to extract the key features. These features and the raw data together are then used to train
and evaluate various ML models, including support vector machines (SVM), convolutional neural networks
(CNN), and long short-term memory (LSTM). Metrics such as accuracy, F1 score, and receiver operating
characteristic (ROC) are employed to evaluate the performance of our approach for AFib detection. Through rigorous experimentation, we identify the combination of features and ML models that are promising in quick and reliable AFib detection. By comparing related works on PPG, we also investigate whether PPG is viable in aiding or replacing ECG as a data source for AFib detection.
Date Acceptance
2026-04-14
Citation
IEEE Access
ISSN
2169-3536
Publisher
IEEE
Journal / Book Title
IEEE Access
Copyright Statement
Copyright This paper is embargoed until publication. Once published the Version of Record (VoR) will be available on immediate open access.
License URL
Publication Status
Accepted
Date Publish Online
2026-04-16