Adaptive Clustering and Incentive Mechanism for Federated Learning in IoT

Federated Learning (FL) enables edge devices to collaboratively train machine learning models while preserving data privacy. However, in IoT networks, constrained devices face challenges such as high prediction errors due to limited dataset sizes and the computational costs associated with FL tasks. To address these issues, this paper proposes a two-level resource management framework for IoT Federated Learning. The first level focuses on forming homogeneous learning clusters with mandatory minimal dataset size where devices with similar computational power and data distribution are grouped together to optimize learning performance. The second level employs a two-stage Stackelberg game to incentivize devices to contribute larger datasets by offering monetary rewards, balancing the trade-off between prediction accuracy, computational costs, and energy consumption. Our framework shows significant improvements in prediction accuracy and overall cost reduction compared to the flat network approach, where all devices are grouped together.