ECG-based heart arrhythmia classification using feature engineering and a hybrid stacked machine learning.

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Heart arrhythmias are irregular heart rhythms with increasing mortality rates.
• Early detection and management of arrhythmias are crucial for improving survival.
• Electrocardiogram (ECG) is the standard diagnostic tool, but expert analysis is time-consuming.

Purpose of the Study:

• To develop and evaluate a hybrid stack classifier model for automated heart arrhythmia classification from ECG signals.
• To compare the performance of the proposed model against conventional and other ensemble machine learning algorithms.
• To investigate the impact of feature selection techniques on classification accuracy.

Main Methods:

• Development of a hybrid stack classifier model using ensemble machine learning techniques.
• Feature engineering using Principal Component Analysis (PCA), Chi-Square, and Recursive Feature Elimination (RFE) to select 50, 65, 80, or 95 features.
• Training and evaluation of various classifiers, including conventional, bagging, boosting, and stack classifiers.
• Utilizing XGBoost as the meta-classifier in the proposed stack classifier model.

Main Results:

• The proposed stack classifier with XGBoost as the meta-classifier achieved the highest performance.
• The model trained with 65 features selected by PCA demonstrated superior results.
• Achieved exceptional performance metrics: 99.58% accuracy, 99.57% precision, 99.58% recall, and 99.57% F1-score.

Conclusions:

• The developed hybrid stack classifier model shows significant promise for accurate and automated arrhythmia diagnosis.
• This automated approach can reduce the reliance on extensive human intervention in ECG analysis.
• The findings suggest a potential for improved patient outcomes through early and precise arrhythmia detection.