A Hybrid Generic Framework for Heart Problem Diagnosis Based on a Machine Learning Paradigm.

Area of Science:

• Cardiology
• Artificial Intelligence
• Machine Learning

Background:

• Accurate heart problem prediction is crucial for reducing mortality and improving patient outcomes.
• Existing machine learning models face challenges due to data heterogeneity across different healthcare systems.
• A need exists for a generalized, robust framework for heart problem identification.

Purpose of the Study:

• To develop a generic, hybrid machine learning framework for reliable heart problem diagnosis.
• To enhance prediction accuracy and minimize diagnostic bias through a novel voting mechanism.
• To demonstrate the framework's adaptability to diverse datasets and measurement variations.

Main Methods:

• A two-layer hybrid framework combining multiple machine learning and deep learning models.
• Implementation of a novel voting technique for consolidating model outputs and reducing bias.
• Integration of a feature selection framework to identify and utilize the most relevant predictive features.

Main Results:

• The proposed framework achieved 95.6% accuracy in heart problem prediction.
• Demonstrated superior performance compared to single machine learning models, classical stacking, and traditional voting techniques.
• Validated the framework's ability to be retrained for datasets with different distributions and measurements.

Conclusions:

• The novel hybrid framework effectively enhances heart problem prediction accuracy and reliability.
• The proposed layered approach with advanced voting neutralizes model bias, offering a more dependable diagnostic tool.
• The framework's generalizability and adaptability make it suitable for diverse clinical applications.