<p>Heart disease is a major cause of death globally, making early and precise diagnosis an essential area of research. The advent of Internet of Medical Things (IoMT) enables continuous patient monitoring making it easier to collect vital sensor data, the machine learning (ML) algorithms have shown better performance for precise diagnosis of patients’ heart conditions. However, conventional ML approaches rely on centralized data collection, which raises significant privacy and security concerns. To address data privacy concerns and enhance model training in a decentralized fashion, federated learning (FL) was introduced. FL works in a decentralized way by enabling the on-device model training on local data while ensuring that the patient’s data never leaves the local device. In this way, FL aligns healthcare systems with stringent data privacy regulations and enhances the privacy of healthcare systems. This survey provides a detailed review of ML and FL techniques for heart disease prediction in edge-enabled IoMT environments. We first develop a taxonomy of methods that includes traditional ML techniques, deep learning methods, and FL-based frameworks. We then conduct a comparative analysis of these techniques, emphasizing their strengths, limitations, and suitability for real-world healthcare scenarios. Additionally, we examine commonly used benchmark datasets for heart disease diagnosis and evaluate their appropriateness for ML and FL applications in IoMT settings. Our findings suggest that FL offers privacy-preserving capabilities while achieving predictive accuracy comparable to, or better than, that of centralized ML methods. We also highlight the key open research challenges of integrating FL-based methods in healthcare systems for heart disease prediction.</p>