Machine learning analysis reveals higher heart disease odds in high-cholesterol men compared to women, highlighting the utility of sex-specific screening.
Key Points
To develop and evaluate supervised machine learning models for predicting heart disease risk from routine clinical variables and test whether high-cholesterol men have higher disease odds than women.
Analyzed a clinical dataset of N=918 patients following the CRISP-DM framework with median imputation for irregular cholesterol values.
Performed feature selection using principal component analysis (PCA) and SelectKBest.
Trained and tuned Logistic Regression, Support Vector Machine (SVM), and a soft voting ensemble using grid search with stratified 5-fold cross-validation.
The soft voting ensemble achieved the highest overall performance (accuracy = 0.940, F1 = 0.950, ROC-AUC = 0.958), with Logistic Regression showing similar diagnostic capability (accuracy = 0.929, F1 = 0.940, ROC-AUC = 0.958).
Among patients with cholesterol ≥240 mg/dL, male sex was an independent, statistically significant predictor of heart disease after adjusting for other clinical variables.