Managing Uncertainty in Federated Learning via Interval Fuzzy Sets and Entropy–Based Fusion

This paper introduces a federated learning framework designed to improve the reliability of diagnostic models under conditions of uncertainty, with a particular focus on medical applications such as breast cancer diagnosis. The proposed method integrates interval-valued fuzzy sets to capture data imprecision and employs logistic regression enhanced with interval-based parameter estimation. Model parameters are aggregated across clients using the Choquet integral, extended with an entropy-based weighting scheme that accounts for both model performance and uncertainty. Experimental results on the Wisconsin breast cancer dataset demonstrate that the proposed federated architecture achieves superior performance compared to traditional methods, particularly in non-IID and unbalanced data scenarios. The framework offers robust privacy preservation, effective uncertainty modeling, and improved classification accuracy, making it suitable for high-stakes, privacy-sensitive domains.