Revolutionizing cardiovascular disease classification through machine learning and statistical methods

Area of Science:

• Medical Informatics
• Computational Biology
• Artificial Intelligence

Background:

• Cardiovascular diseases (CVDs) encompass a range of heart and blood vessel conditions.
• Traditional CVD diagnosis relies on expensive and time-consuming expert clinical evaluation.
• Emerging machine learning (ML) and statistical techniques enable cost-effective digital CVD diagnosis.

Purpose of the Study:

• To classify cardiovascular diseases (CVDs) using 19 machine learning (ML) models.
• To evaluate and rank the performance of ML models for CVD classification.
• To assess the efficiency and reliability of ML models using benchmark datasets.

Main Methods:

• Utilized 19 ML models for CVD classification.
• Employed two benchmark CVD datasets from Kaggle and UCI repositories.
• Performed 50 simulations for each model and dataset, applying nonparametric statistical tests.

Main Results:

• Extra Tree Classifier demonstrated statistically superior accuracy and precision.
• Extreme Gradient Boost (XGBoost) classifier achieved statistically superior recall, kappa, and F1 scores.
• XGBRF classifier ranked second for recall performance.

Conclusions:

• ML models provide a viable alternative for CVD diagnosis.
• Specific ML models show distinct strengths in different performance metrics for CVD classification.
• Statistical testing confirms the significant performance differences among evaluated ML models.