LSTM and GRU based Traffic Prediction Using Live Network Data

Abstract

Artificial Intelligence can help 5G and Beyond networks to manage the complexity as a result of integration of different technologies and the design requirement specific to vertical needs. With this paradigm, it is essential to predict the network metrics in order to be prepared for future demand in the network. In this paper, we have focused on RNN based traffic volume prediction of an eNodeB. We use live network metrics to train our multivariable forcasting model to analyze the traffic demand and predict the network traffic by using LSTM and GRU models. Our results show the importance of multi-variable approach and that both models can handle the demand increase in busy hours of days when the traffic load is in line with its periodical trend. Sudden increases or decreases in traffic load may sometimes create problem such as prediction of more or less resources than required. GRU performs better to handle sudden increase or decrease situations.