ECG arrhythmia classification via wavelet-driven feature extraction and swarm-optimised gradient boosting.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Signal Processing

Background:

• Cardiovascular diseases are a leading cause of global mortality.
• Accurate and efficient detection of arrhythmia from Electrocardiogram (ECG) signals is critical.
• Existing methods face challenges with class imbalance in ECG datasets.

Purpose of the Study:

• To develop a lightweight and computationally efficient framework for ECG beat analysis.
• To improve the accuracy and efficiency of arrhythmia detection.
• To address the challenge of severe class imbalance in the MIT-BIH Arrhythmia Database.

Main Methods:

• Utilized Discrete Wavelet Transform (DWT) for feature extraction.
• Combined DWT-based statistical features with ECG morphological descriptors.
• Employed Artificial Bee Colony (ABC) algorithm to optimize eXtreme Gradient Boosting Machine (XGBM) hyperparameters.
• Pre-processed ECG signals using a seven-stage algorithm including filtering and R-peak detection.

Main Results:

• The proposed ABC-XGBM model achieved 95.14% classification accuracy and a macro F1-score of 0.948.
• Discrete Wavelet Transform (DWT) improved accuracy by +3.7%, and ABC optimization by +1.14%.
• Demonstrated stable performance with a mean accuracy of 0.952 ± 0.001 via five-fold cross-validation.
• Outperformed deep learning models like CardioAttentionNet (91.20%) and transformer-based classifiers (90.50%).

Conclusions:

• The proposed framework offers a precise and efficient solution for ECG beat analysis and arrhythmia detection.
• The integration of DWT features and ABC-optimized XGBM effectively handles class imbalance.
• The computationally efficient design makes it suitable for real-time applications without GPU dependency.