A privacy-preserving rule fusion approach for uncertainty-aware decision-making in posture detection

This study introduces a novel approach to constructing rule-based decision systems within a federated learning framework. The proposed rule fusion method is specifically designed for privacy-sensitive applications with uncertainty in data environments. Such challenges frequently arise in medical and social applications, such as fall detection in elderly care facilities, where datasets are often limited, imbalanced, or noisy and strict data privacy regulations prevent traditional data-sharing practices. Federated learning offers a practical solution by allowing multiple institutions to collaboratively train models without exchanging raw data. Our method extends this concept by enabling the classification of decentralized datasets through the collaborative generation and integration of rule sets, maintaining full data privacy throughout the process. Furthermore, to systematically manage the inherent uncertainties in the data, our approach incorporates interval-valued fuzzy set theory alongside knowledge measures. This allows us to handle uncertainty both in knowledge representation and during reasoning by applying uncertainty-aware measures within AI techniques. We demonstrate the effectiveness of our approach on posture detection, a critical element of elderly care. Our federated, privacy-preserving system not only advances healthcare technologies but also addresses crucial concerns around data confidentiality, offering significant societal and economic benefits.