Enhanced electrocardiogram machine learning-based classification with emphasis on fusion and unknown heartbeat...

Area of Science:

• Cardiology
• Machine Learning
• Biomedical Engineering

Background:

• Early arrhythmia detection is crucial for patient outcomes.
• Electrocardiogram (ECG) analysis is a primary diagnostic tool.
• Classifying diverse heartbeat types, including rare ones, presents a challenge.

Purpose of the Study:

• To develop a reliable ECG heartbeat classification model.
• To enhance the prediction accuracy for uncommon heartbeat types (fusion and unknown beats).
• To compare the performance of five machine learning algorithms for ECG classification.

Main Methods:

• Utilized the MIT-BIH SupraVentricular Database.
• Employed synthetic minority oversampling technique (SMOTE) and data augmentation for rare classes.
• Trained and evaluated logistic regression, Random Forest (RF), K-nearest neighbor, linear support vector machine, and linear discriminant analysis models.

Main Results:

• The Random Forest (RF) algorithm demonstrated superior performance.
• Achieved an overall accuracy of 97%.
• Significantly improved recall for rare fusion (F) and unknown (Q) beats compared to existing literature, with specific recall values of 97% (N), 93% (SVEB), 95% (VEB), 95% (F), and 30% (Q).

Conclusions:

• The developed RF model offers a reliable solution for ECG heartbeat classification.
• The data augmentation and SMOTE techniques effectively addressed the challenge of classifying rare heartbeats.
• This approach holds promise for improving early arrhythmia detection through enhanced ECG analysis.