MACHINE LEARNING TECHNIQUES FOR CARDIOVASCULAR DISEASE PREDICTION

Abstract

Cardiovascular disease (CVD) continues to be the leading cause of death globally, accounting for millions of lives annually. Early and accurate prediction of CVD is critical for prevention and intervention, yet existing clinical approaches often fail to capture the complex interactions among multiple risk factors. While machine learning (ML) has been increasingly applied to CVD prediction, many studies rely on single datasets, limited sample sizes, or overlook systematic evaluation across different algorithms. This study addresses these limitations by employing a merged dataset of 920 patient records, created by integrating four well-known heart disease datasets from Cleveland, Hungarian, Switzerland, and Long Beach VA based on 11 common features. The integration enhances data diversity and improves model generalization compared to single-dataset approaches. Four supervised ML algorithms, namely, Logistic Regression, Naïve Bayes, Decision Tree, and K-Nearest Neighbour (KNN), were systematically implemented and evaluated. Data preprocessing involved feature scaling for non-tree-based algorithms and tailored encoding strategies for categorical variables. Model training and validation were carried out using Stratified 5-Fold Cross-Validation, ensuring balanced representation of classes across folds. The results shows KNN algorithm consistently outperformed all models, achieving 91.6% accuracy with precision and recall of 92%, highlighting its robustness in capturing non-linear decision boundaries in heterogeneous datasets. Importantly, the study shows that integrating multiple datasets strengthens predictive power and improves the reliability of ML models in cardiovascular diagnosis. Findings demonstrate that simple yet effective models such as KNN, when applied to enriched datasets, can deliver clinically meaningful insights and serve as practical tools for early risk detection.