Intra-Cluster Federated Learning-Based Model Transfer Framework for Traffic Prediction in Core Network

Abstract

Accurate prediction of cellular traffic will contribute to efficient operations and management of mobile network. With deep learning, many studies have achieved exact cellular traffic prediction. However, the reality is that quite a few subnets in the core network do not have sufficient computing power to train their deep learning model, which we call subnets (LCP-Nets) with limited computing power. In order to improve the traffic prediction efficiency of LCP-Nets with the help of deep learning and the subnets (ACP-Nets) with abundant computing power under the requirement of privacy protection, this paper proposes an intra-cluster federated learning-based model transfer framework. This framework customizes models for LCP-Nets, leveraging transferring models trained by ACP-Nets. Experimental results on the public dataset show that the framework can improve the efficiency of LCP-Nets traffic prediction.