Real-time ECG-based detection of cardiovascular diseases using balanced and interpretable machine learning approaches

Area of Science:

• Cardiology and Medical Informatics
• Machine Learning in Healthcare
• Signal Processing for Biomedical Applications

Background:

• Cardiovascular diseases (CVDs) remain the leading global cause of mortality.
• There is a critical need for reliable and accurate diagnostic systems for CVDs.
• Existing electrocardiogram (ECG) datasets may lack standardization or sufficient diversity for robust machine learning applications.

Purpose of the Study:

• To create a standardized, large-scale 12-lead ECG dataset for classifying six major cardiovascular diseases.
• To evaluate the efficacy of various machine learning and deep learning models for CVD detection.
• To investigate feature selection and extraction techniques, including the impact of addressing class imbalance using SMOTE.

Main Methods:

• Collected and pre-processed a dataset of 34,580 12-lead ECG recordings from a clinical setting.
• Extracted 14 clinically informative features and applied the Synthetic Minority Oversampling Technique (SMOTE) to balance the dataset.
• Trained and tested ten machine learning and deep learning models (e.g., Random Forest, DNN, MLP, RNN) and employed SHAP/LIME for interpretability.

Main Results:

• On the raw dataset, Random Forest and Gradient Boosting achieved near-perfect performance (99.88% accuracy, precision, recall, F1-score).
• After SMOTE, deep learning models showed significant improvements: DNN (97.62% accuracy) and MLP (98.49% F1-score), demonstrating enhanced generalization.
• The balanced dataset and SMOTE application proved effective in improving model stability and performance, especially for deep networks.

Conclusions:

• The developed standardized ECG dataset and machine learning models provide a highly accurate and interpretable tool for real-time CVD classification.
• SMOTE is a crucial technique for mitigating class imbalance, significantly boosting the performance of deep learning models in cardiovascular diagnostics.
• The findings support the clinical utility of the proposed system for decision support in diagnosing cardiovascular diseases.