Federated Learning–Driven Trustful and Privacy-Preserving Service Composition in Virtual Hospitals

Abstract

: The increasing adoption of virtual hospitals and Internet of Medical Things (IoMT)-driven clinical workflows demands service composition mechanisms that can operate reliably under strict privacy constraints and heterogeneous provider ecosystems. Traditional trust-based composition approaches, such as majority voting or Bayesian aggregation, struggle to remain robust in large-scale, non-Independent and Identically Distributed (IID), and privacy-sensitive environments. This paper addresses these limitations by introducing PSCP-FL, a Federated Learning– Driven Privacy-and Trust-Aware Service Composition Framework for cloud-and edge-enabled virtual hospitals. We first define a comprehensive set of medical service trust attributes and construct a Federated Learning–based Trust Network (FLTN) that predicts provider trust while preserving data locality. FLTN enables the composer to prune untrustworthy services and select optimal compositions using a federated, dynamically updated evaluation model. Experiments demonstrate that PSCP-FL outperforms the Majority Voting and Bayesian Average-Based Trustful Service Composition (MVBA-TSC) baseline in standard deviation reduction (up to 40%), trust accuracy, robustness under non-IID and adversarial conditions, and execution time (15%–30% faster), while producing more stable and homogeneous service chains. These results confirm that PSCP-FL provides a reliable and privacy-preserving foundation for the composition of critical virtual