(EXTREME GRADIENT BOOSTING) XGBOOST AND MACHINE LEARNING-BASED EARLY DETECTION OF CORONARY ARTERY DISEASE USING CLINICAL DATA

Abstract

The number one cause of death in the world is still coronary artery disease. Early detection is necessary to provide a proper treatment plan that will improve the outcome of the patient. Many times, traditional diagnostic tests are invasive or fail to detect the early stages of (Coronary Artery Disease) CAD. This research paper is designed to create and evaluate machine learning models, specifically (Extreme Gradient Boosting) XGBoost, for the purpose of detecting early-stage CAD utilizing clinical data. The performance of these machine learning models will be compared to traditional diagnostic techniques. Clinical data on 1025 patients were reviewed. These included demographic information, lab results, vital signs, and the medical history of each patient. The development of numerous machine learning algorithms was completed and included the creation of Support Vector Machine, Random Forest, Logistic Regression and (Extreme Gradient Boosting) XGBoost. In order to evaluate each of the models created, accuracy, sensitivity, specificity and the area under the receiver operator characteristic curve (ROC) were utilized. Accuracy was 94.2%, Sensitivity was 92.8%, Specificity was 95.1% and Area Under Curve was 0.967 for the (Extreme Gradient Boosting) XGBoost Model. Age, Cholesterol Level, Type of Chest Pain and Angina During Physical Activity were identified as the Top Four Predictive Factors of CAD. The (Extreme Gradient Boosting) XGBoost Model was able to achieve a greater degree of accuracy compared to the traditional risk calculator for assessment of a patient's risk of developing CAD. The (Extreme Gradient Boosting) XGBoost Model produced 18.5% greater accuracy compared to the traditional risk calculator. Use of Machine Learning Models, particularly the (Extreme Gradient Boosting) XGBoost Model, may provide clinicians with the ability to perform early detection of CAD from Clinical Data. Non-invasive, Early Screening Techniques may allow clinicians to make better treatment decisions regarding High Risk Patients.