A Systematic Review of Machine Learning Techniques for Cardiovascular Disease Prediction With Performance Analysis and Future Perspectives

Cardiovascular disease is a chronic and significant health issue in the world, highlighting the need to detect and diagnose it in its early stages to increase the survival chances of the patient. The use of machine learning, as well as deep learning, has grown rapidly in medical research and shows great potential for improving the accuracy of heart disease prediction. They are able to identify latent connections between various clinical, demographic, and lifestyle variables and can be more accurate than traditional statistical techniques. This review follows a systematic literature review approach based on PRISMA guidelines, ensuring a structured and transparent selection of relevant studies. This review discusses some of the typical supervised learning algorithms used to assess cardiovascular risk, including Random Forests, Logistic Regression, Decision Trees, Neural Networks, Gradient Boosting, and Support Vector Machine. The heart disease dataset, the Cleveland and Framingham Heart Study datasets, and the Kaggle datasets are the most popular datasets used in the studies. Model performance is evaluated using metrics such as precision, recall, F1-score, and area under the receiver operating characteristic curve; however, in medical diagnosis, recall and F1-score are particularly important to minimize false negatives. Although these approaches show promising results, challenges such as imbalanced datasets, limited interpretability, and poor generalization across diverse populations remain. Furthermore, variations in datasets and experimental setups limit direct comparison across studies. This review highlights current research trends, compares existing methods, and identifies key limitations. It also provides recommendations for future research to develop more reliable, interpretable, and clinically applicable heart disease prediction systems.