A systematic review of machine learning in heart disease prediction

Background/aim: Cardiovascular diseases (CVDs) are a leading cause of global mortality, prompting the need for advanced predictive tools. While machine learning (ML) offers a powerful solution, there are significant challenges to clinical translation. This systematic review synthesizes the current state of ML in heart disease prediction, evaluating algorithmic performance, data utilization, and key translational challenges. Materials and methods: Following PRISMA guidelines, a systematic search of literature published up to 2025 was conducted. From an initial pool of over 2500 records, a rigorous screening process yielded 65 studies for in-depth qualitative synthesis. Results: Analysis showed that ensemble learning models dominate prediction tasks on structured data, achieving high accuracy on benchmarks. Deep learning (DL) is increasingly applied to unstructured data like electrocardiogram signals and cardiac imaging. Despite high performance reported in models, a significant translational gap exists. This is driven by a pervasive lack of external validation, an overreliance on limited public datasets, and the black-box nature of complex models that reduces clinical trust. The adoption of explainable artificial intelligence is a key trend aimed at mitigating these challenges. Conclusion: While ML shows significant potential, its utility remains largely confined to academic settings. The future of the field depends on a fundamental research shift, rather than on incremental accuracy gains. Progress requires a concerted focus on robust external validation, the development of large-scale representative datasets, and the creation of interpretable systems that can be effectively integrated into clinical workflows to improve patient outcomes.