A Metaheuristic Optimization Approach for Parameter Estimation in Arrhythmia Classification from Unbalanced Data.

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Electrocardiogram (ECG) analysis is crucial for detecting heart diseases through bio-signal abnormality classification.
• Classifying arrhythmias from unbalanced ECG data presents significant challenges in model optimization and computational resource allocation.
• Existing methods often rely on empirical parameter tuning rather than advanced optimization techniques.

Purpose of the Study:

• To introduce a metaheuristic optimization approach for parameter estimation in arrhythmia classification using unbalanced ECG data.
• To enhance the accuracy, sensitivity, and precision of ECG-based arrhythmia detection.
• To address the complexities of classifying imbalanced datasets in cardiac bio-signal analysis.

Main Methods:

• Utilized an unbalanced subset of ECG databases for classifying eight types of arrhythmia.
• Implemented a hybrid approach combining data-level undersampling (clustering-based) and algorithmic-level feature selection.
• Compared two metaheuristic optimization algorithms: differential evolution and particle swarm optimization for parameter estimation.

Main Results:

• Achieved high classification performance metrics on unbalanced data: 99.95% accuracy, 99.88% F1 score, 99.87% sensitivity, 99.89% precision, and 99.99% specificity.
• Demonstrated the effectiveness of the metaheuristic optimization in improving ECG classification accuracy.
• Successfully tackled the challenge of imbalanced classes in arrhythmia detection.

Conclusions:

• The proposed metaheuristic optimization approach significantly enhances arrhythmia classification from unbalanced ECG data.
• The combined data and algorithmic level strategies effectively manage imbalanced datasets.
• This method offers a robust solution for accurate and sensitive heart bio-signal abnormality classification.