A Novel Group Management Scheme of Clustered Federated Learning for Mobile Traffic Prediction in Mobile Edge Computing Systems

This study developed a novel group management scheme based on clustered federated learning (FL) for mobile traffic prediction (referred to as FedGM) in mobile edge computing (MEC) systems. In FedGM, to improve the convergence time during the FL procedure, we considered multiple MEC servers to first be clustered based on their geographic locations and augmented data patterns as references for clustering. In each cluster, by alleviating the straggler impact owing to the heterogeneity of MEC servers, we then designed a group management scheme that optimizes i) the number of groups to be created and ii) the group association of the MEC servers by minimizing their average idle time and group creation cost. For this purpose, we rigorously formulated analytical models for the computation time for local training and estimated the average idle time by applying different frequencies of local training over the MEC servers. The optimization problem was designed using a nonconvex problem, and thus a genetic-based heuristic approach was devised for determining a suboptimal solution. By reducing the average idle time, thereby increasing the workload of the MEC servers, the experimental results for two real-world mobile traffic datasets show that FedGM surpasses previous state-of-theart methods in terms of convergence speed with an acceptable accuracy loss.