Hybrid machine learning models for enhanced arrhythmia detection from ECG signals using autoencoder and convolution...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Early detection of cardiac disease (CD) is vital for timely treatment.
• Electrocardiogram (ECG) signals are crucial for identifying arrhythmias.
• Automated arrhythmia detection systems can aid in large-scale healthcare screening.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for improved ECG arrhythmia detection.
• To compare the performance of autoencoder and convolutional feature extraction schemes.
• To identify the best-performing ML models for real-time arrhythmia detection.

Main Methods:

• Developed eight ML models using two feature extraction schemes: autoencoder and convolutional.
• Trained and tested models on MIT-BIH Arrhythmia and ECG 5000 datasets.
• Utilized TOPSIS and mRMR for ranking ML models and identifying top performers.

Main Results:

• Models using autoencoder features demonstrated superior performance over convolutional features.
• The hybrid Autoencoder Features with Neural Network (AEFNN) model achieved 97.96% accuracy on the MIT-BIH dataset.
• The AEFNN model achieved 99.20% accuracy on the ECG 5000 dataset.

Conclusions:

• The proposed AEFNN model is effective for accurate and early arrhythmia detection.
• Autoencoder-based features enhance ML model performance for ECG analysis.
• This approach can support timely diagnosis and intervention in cardiac disease management.