Federated Learning for Network Traffic Prediction

Network traffic prediction is imperative for effective network planning, decision-making, and optimization, leveraging the inherent predictability observed in traffic patterns. Although various machine learning models, such as deep neural networks (NNs) with recurrent units, including long-short-term memory (LSTM) NNs, have shown promise in this area, they are conventionally centralized trained, overlooking the potential benefits of distributed training techniques at the network edge. Exploiting advances in distributed training could provide certain advantages to the problem at hand, such as improved privacy preservation, reduced processing times, and lower bandwidth usage. Thus, this work proposes a federated learning (FL) framework for predicting network traffic, utilizing LSTMs for local model training. Simulation results validate the effectiveness of the proposed model, as this distributed implementation demonstrates high traffic prediction accuracy on real traffic traces, comparable to the traditional centralized approach, while safeguarding data privacy.