A prototype-assisted clustered federated learning for big data security and privacy preservation

Yalan Jiang, Dan Wang, Bin Song, Xiaojiang Du

Abstract: Highlights•Introducing a Clustered FL approach to enhance privacy and security in handling IoT big data.•Addressing CFL challenges like multi-distribution data, cluster similarity, and class imbalance.•MDSPFL allows local datasets to follow multiple distributions, considering class imbalance.•MDSPEL handles workload surges and enforces uniform feature space distribution using prototypes.•Experiments show MDSPFL outperforms FedAvg, FedPer, and FedNH in multi-distribution, imbalanced settings.