A comprehensive machine learning approach for early detection of cardiovascular diseases using machine learning techniques

Cardiovascular disease (CVD) remains the leading global cause of death, accounting for approximately 17.9 million deaths annually. Using the Framingham Heart Study dataset, this study assesses optimized machine learning techniques for CVD prediction. Class imbalance was addressed by a thorough preprocessing pipeline that included borderline-SMOTE2, partial record elimination, and feature standardisation. Accuracy, ROC-AUC, sensitivity, specificity, F1-score, and Cohen’s Kappa were used to assess four highly optimized classifiers: Random Forest, AdaBoost, Support Vector Machine (SVM), and Decision Tree. With 94.28% accuracy, a ROC-AUC of 0.9783, sensitivity of 0.9371, specificity of 0.9485, F1-score of 0.9424, and Cohen’s Kappa of 0.8856, AdaBoost produced the best results. SVM demonstrated high sensitivity (0.9419) but low specificity (0.8631), but the decision tree did not perform well. Results confirm that ensemble-based approaches provide superior stability, balanced classification, and better generalisation for cardiovascular risk prediction. The proposed framework offers a reliable, interpretable, and clinically applicable decision-support solution for early CVD detection.