Mobile Traffic Prediction at the Edge Through Distributed and Deep Transfer Learning

TL;DR

This work tackles privacy-preserving, energy-efficient mobile traffic prediction by deploying DL at the network edge and leveraging Deep Transfer Learning to share knowledge across base stations without central data aggregation. It compares CNN- and RNN-based predictors trained stand-alone and via DTL on LTE control-channel data from Barcelona, showing CNNs generally achieve higher accuracy while DTL improves performance in the majority of cases and dramatically reduces training energy. The authors quantify complexity and energy savings, observing up to ~60% energy reduction for CNNs and ~90% for RNNs under DTL, and they apply two XAI methods (SmoothGrad and LRP) to interpret learned models. The results support edge-native, privacy-preserving, and energy-conscious deployment for real-time network management, with future work focusing on multi-task ensembles and deeper XAI-assisted optimization.

Abstract

Traffic prediction represents one of the crucial tasks for smartly optimizing the mobile network. Recently, Artificial Intelligence (AI) has attracted attention to solve this problem thanks to its ability in cognizing the state of the mobile network and make intelligent decisions. Research on this topic has concentrated on making predictions in a centralized fashion, i.e., by collecting data from the different network elements and process them in a cloud center. This translates into inefficiencies due to the large amount of data transmissions and computations required, leading to high energy consumption. In this work, we investigate a fully decentralized AI solution for mobile traffic prediction that allows data to be kept locally, reducing energy consumption through collaboration among the base station sites. To do so, we propose a novel prediction framework based on edge computing and Deep Transfer Learning (DTL) techniques, using datasets obtained at the edge through a large measurement campaign. Two main Deep Learning architectures are designed based on Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) and tested under different training conditions. Simulation results show that the CNN architectures outperform the RNNs in accuracy and consume less energy. In both scenarios, DTL contributes to an accuracy enhancement in 85% of the examined cases compared to their stand-alone counterparts. Additionally, DTL significantly reduces computational complexity and energy consumption during training, resulting in a reduction of the energy footprint by 60% for CNNs and 90% for RNNs. Finally, two cutting-edge eXplainable Artificial Intelligence techniques are employed to interpret the derived learning models.

Figures (10)

• Figure 1: Example of the Edge AI scenario in mobile networks.
• Figure 2: Average Pearson correlation matrix across the three datasets, each containing the same five variables. This matrix represents the average pairwise Pearson correlation coefficients calculated for the three datasets.
• Figure 3: Variables time series on a sample of 7 days of the PS dataset for RNTI$_{count}$ (a), RB$_{down}$ (b) and RB$_{up}$ (c). Each color corresponds to a day of the week, starting from Monday which is the first represented with blue color.
• Figure 4: THR$_{down}$ and THR$_{up}$ time series by date for the three datasets on a sample of 7 days. The shared y-axis unit of measurement is billions of bits.
• Figure 5: RNN model's structure graphical representation.