Heart Disease Prediction Models Performance Analysis based on Logistic Regression, Random Forest and XGBoost

Abstract

Heart disease is a leading cause of mortality worldwide, underscoring the urgent need for effective early detection techniques. This study evaluates the predictive effectiveness of three machine learning models-Logistic Regression, Random Forest, and XGBoost-using data from the CDC’s Behavioral Risk Factor Surveillance System (BRFSS), focusing on individuals aged 60–69. The class weighting method was employed to address class imbalance. The dataset was partitioned into training and testing sets in a 7:3 ratio. The assessment of model performance was performed using accuracy, sensitivity, specificity, F1-score, and the area under the ROC curve (AUC). The Random Forest model achieved the highest sensitivity (0.81) and F1-score (0.87), effectively identifying affirmative cases, however it demonstrated lower specificity (0.51). XGBoost demonstrated a balanced accuracy of 0.73, with a sensitivity of 0.75 and a specificity of 0.58. Logistic Regression had the highest AUC (0.77) and specificity (0.67), along with notable sensitivity (0.72), offering the ideal equilibrium between true positive and false positive rates. Logistic Regression was selected as the final model because of its equitable performance and clinical interpretability. These findings underscore the importance of interpretable models in medical applications and advocate for their integration into preventative cardiovascular care. Future research should encompass more extensive clinical characteristics and validate findings across diverse populations.