An improved electrocardiogram arrhythmia classification performance with feature optimization

Area of Science:

• Cardiology
• Biomedical Engineering
• Data Science

Background:

• Electrocardiography (ECG) data analysis for arrhythmia classification faces challenges due to large datasets and numerous potential features.
• Increasing complexity of clinical symptoms necessitates efficient feature identification to avoid misclassification.

Purpose of the Study:

• To identify optimal features for accurate ECG arrhythmia classification.
• To evaluate shallow and deep feature extraction techniques combined with metaheuristic optimization.

Main Methods:

• Applied shallow and deep feature extraction techniques to ECG signals.
• Utilized a metaheuristic optimization algorithm for feature selection post-extraction.

Main Results:

• Shallow feature extraction (time-domain analysis) with metaheuristic optimization outperformed other methods.
• Selection of 1-3 RR-interval features achieved 100% accuracy, sensitivity, specificity, and precision.

Conclusions:

• The proposed end-to-end architecture is simple and has low complexity.
• This approach is highly effective for practical ECG arrhythmia classification applications.