The rapid increase in the number of internet of things (IoT) devices in the healthcare sector has produced an ecosystem of multimodal data that includes physiological signals, behavioral measurements, and electronic health records. Nevertheless, centralized analytics are dangerous to the privacy and safety of patient data. Federated learning (FL) has become a prospective paradigm of decentralized intelligence, which allows the joint training of models without providing actual data. The review is a systematic review of research published during the 2020–2025 period on privacy‐preserving federated recommender systems for healthcare applications in IoT. With the PRISMA protocol, the studies of interest are categorized under five thematic aspects: (i) privacy preservation techniques including differential privacy, secure aggregation, and homomorphic encryption, (ii) personalization and model optimization approaches, (iii) multimodal fusion systems, (iv) communication efficiency systems, and (v) metrics of evaluation in real‐time implementation. As the analysis indicates, hybrid privacy models and adaptive optimization can make a major contribution to the improvement of personalization and scalability, but the issue remains in the manner of providing efficient communication and fairness between heterogeneous clients. In addition, the paper identifies open research directions, including quantum‐resilient security and the integration of edge intelligence and sustainability‐monitoring model training. The results of this review provide a single taxonomy, benchmarking standards, and a research direction for future studies to improve trustworthy, real‐time, and privacy‐compliant recommender systems in healthcare IoT settings.
N. et al. (Thu,) studied this question.