A study on heart data analysis and prediction using advanced machine learning methods

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Cardiovascular diseases are a leading global cause of mortality.
• Early detection and accurate diagnosis are vital for preventing severe outcomes like heart attacks.
• Patient-centered systems require reliable predictive models for high-risk individuals.

Purpose of the Study:

• To compare the performance of classic machine learning (ClassicML) and automated machine learning (AutoML) models for early cardiovascular disease prediction.
• To identify the optimal machine learning approach for developing patient-centered systems for heart attack risk assessment.
• To evaluate the impact of feature engineering and data balancing techniques on model performance.

Main Methods:

• Utilized a combined dataset from four universities (Swiss, Hungarian, Cleveland, Long Beach VA) with 12 key features.
• Implemented and compared nine traditional machine learning algorithms against seven automated machine learning algorithms.
• Assessed model performance using accuracy, F1 score, precision, and recall.

Main Results:

• Automated machine learning (AutoML) tools are not consistently superior to traditional machine learning (ClassicML) methods.
• Effective feature extraction, appropriate data balancing, and a suitable machine learning model are critical for optimal performance.
• Specific combinations of techniques yielded the best predictive accuracy for cardiovascular events.

Conclusions:

• The choice of machine learning methodology (ClassicML vs. AutoML) should be carefully considered based on specific application needs.
• Feature engineering and data balancing are crucial components for enhancing the predictive power of cardiovascular disease models.
• Further research into optimized feature selection and data preprocessing can improve patient-centered predictive systems.