Imbalanced ECG signal-based heart disease classification using ensemble machine learning technique.

Area of Science:

• Cardiology
• Machine Learning
• Signal Processing

Background:

• Machine learning (ML) models are increasingly used for automated heart disease (HD) detection via physiological signals like ECG.
• Electrocardiogram (ECG)-based detection is the most prevalent clinical method for identifying heart conditions.
• This study addresses the challenge of using imbalanced ECG datasets for training classification models.

Purpose of the Study:

• To evaluate the efficacy of rarely used ML classifiers (SVM, LR, AdaBoost) for heart disease detection.
• To develop and assess an ensemble model combining top-performing classifiers for improved diagnostic accuracy.
• To validate the performance of the proposed models using established public ECG datasets.

Main Methods:

• Selected and trained Support Vector Machine (SVM), Logistic Regression (LR), and Adaptive Boosting (AdaBoost) models on imbalanced ECG data.
• Evaluated model performance using accuracy, F1-score, and Area Under Curve (AUC) metrics.
• Ensembled the top-performing LR and AdaBoost classifiers using a majority voting principle.

Main Results:

• AdaBoost and Logistic Regression classifiers demonstrated superior performance compared to SVM for heart disease detection.
• The proposed ensemble model achieved high performance metrics: Accuracy (0.946 PTB-ECG, 0.921 MIT-BIH), F1-score (0.949 PTB-ECG, 0.926 MIT-BIH), and AUC (0.951 PTB-ECG, 0.950 MIT-BIH).
• The ensemble approach significantly enhanced the detection capabilities for heart diseases.

Conclusions:

• The developed ensemble ML model offers a robust and accurate method for heart disease detection using ECG signals.
• The methodology shows potential for application with other physiological signals (ICG, MCG, HS) and for detecting different diseases.
• This approach provides a valuable tool for clinicians in diagnosing heart conditions more effectively.