Prediction of atherosclerosis using machine learning based on operations research

Area of Science:

• Cardiovascular Medicine
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Atherosclerosis is a primary cause of cardiovascular diseases like coronary heart disease and cerebral infarction.
• Early intervention by identifying risk factors is crucial for managing atherosclerosis, as early stages lack obvious symptoms.
• Existing machine learning models for atherosclerosis prediction can be hampered by information redundancy from strongly correlated features.

Purpose of the Study:

• To combine statistical analysis and machine learning to reduce feature information redundancy.
• To enhance the accuracy of atherosclerosis diagnosis and prediction systems.

Main Methods:

• Data from a retrospective study was analyzed, initially reducing 34 features to 25 by filtering those with excessive missing values.
• Statistical tests identified 20 distinguishing features, followed by graph theory-inspired machine learning with optimal correlation distances to select 15 significant features.
• Ensemble learning was employed to integrate information from the 5 non-significant features, further improving prediction.

Main Results:

• The initial prediction model achieved an Area Under the ROC Curve (AUC) of 0.84035 and a Kolmogorov-Smirnov (KS) value of 0.646.
• Feature selection using optimal correlation distance improved AUC to 0.88268 and KS to 0.688.
• Ensemble learning further boosted the AUC to 0.89637.

Conclusions:

• The proposed optimal distance feature screening model enhances atherosclerosis prediction accuracy and AUC metrics.
• This approach effectively reduces information redundancy, leading to superior predictive performance.
• The developed code and models are publicly available for further research and application.