An Efficient and Privacy-Preserving AdaBoost Federated Learning Framework for AiP System

As the global population continues to age rapidly, Aging in Place (AiP) solutions have become increasingly vital for enabling elderly individuals to maintain their independence and continue living comfortably in their own homes. These solutions leverage advanced technologies such as smart homes and remote health monitoring. However, in real-world AiP applications, the health data needed for accurate predictions is often spread across multiple medical institutions, which raises signficant privacy concerns when integrating and analyzing the data. To address this challenge, we propose an efficient and privacy-preserving AdaBoost learning framework for vertically partitioned AiP data by utilizing Symmetric Homomorphic Encryption (SHE) technique. To ensure compatibility with the integer-based constraints of SHE, we adopt a straightforward weight quantization strategy by representing AdaBoost sample weights as integers. This design simplifies encrypted computation and maintains the boosting mechanism’s effectiveness. Our theoretical and experimental evaluations validate both the accuracy and security of the proposed framework, highlighting its practical viability for deployment in real-world AiP systems.