A Comparative Study of Machine Learning Models for Cardiovascular Risk Prediction Using PCA-Transformed Framingham Data

Cardiovascular disease (CVD) remains the leading cause of mortality globally, highlighting the urgent need for effective early detection systems. Supervised machine learning algorithms, such as Logistic Regression (LR), SGDClassifier (SDGC), Support Vector Classifier (SVC), K-Nearest Neighbors (KNN), Decision Tree (DT), Random Forest (RF), Gradient Boosting (GB), AdaBoost (AB), LightGBM (LGBM), and XGBoost (XGB), are applied to the Framingham Heart Study (FHS) data set to predict mortality risk and are presented in this research in a comparative analysis. The proposed research includes Principal Component Analysis (PCA) for dimensionality reduction to reduce model dimensionality and improve model efficiency and generalization. The accuracy, precision, recall, F1 score, and AUC-ROC were used as performance metrics for the model, and 10-fold cross validation was used to test the stability of the model. More results show that ensemble models are much better at predictive accuracy and generalization than linear and moderate accuracy classifiers. Test accuracy was 98.05% and F1 score 0.98 for RF and 98.04% for XGB with a similar robustness and greater precision recall balance. PCA is also shown to successfully retain important variance and decrease model complexity in the research. Integration of ensemble learning and dimensionality reduction techniques into wearable compatible, real time cardiovascular monitoring systems helps to scale and accurately provide an early intervention for at risk people, and these findings support it.