Patient-Centric Hierarchical Federated Learning for Heart Rate Prediction

Heart rate (HR) prediction plays a critical role in cardiovascular health monitoring, psychological condition assessment, and athletic performance enhancement. However, utilizing machine learning for HR forecasting poses significant privacy and security challenges. This study introduces a federated learning (FL) method, termed patient-centric hierarchical federated learning (PCHFL), which integrates with differential privacy to protect sensitive health data during HR prediction. The proposed PCHFL method operates across remote home environments where edge devices process data for individual users and hospital-based systems, where data from multiple patients are collected and processed on shared edge devices. By incorporating differential privacy, PCHFL enhances security throughout the FL process. We evaluate the performance of PCHFL against centralized learning and assess its efficacy with differential privacy over 30-second and 60-second windows. The results demonstrate that PCHFL balances privacy, security, and accuracy, providing a scalable solution for healthcare.