AI-Based Traffic Forecasting in 5G Network

Forecasting of the telecommunication traffic is the foundation for enabling intelligent management features as cellular technologies evolve toward fifth-generation (5G) technology. Since a significant number of network slices are deployed over a 5G network, it is crucial to evaluate the resource requirements of each network slice and how they evolve over time. Mobile network carriers should investigate strategies for network optimization and resource allocation due to the steadily increasing mobile traffic. Network management and optimization strategies will be improved if mobile operators know the cellular traffic demand at a specific time and location beforehand. The most effective techniques nowadays devote computing resources in a dynamic manner based on mobile traffic prediction by machine learning techniques. However, the accuracy of the predictive models is critically important. In this work, we concentrate on forecasting the cellular traffic for the following 24 hours by employing temporal and spatiotemporal techniques, with the goal of improving the efficiency and accuracy of mobile traffic prediction. In fact, a set of real-world mobile traffic data is used to assess the efficacy of multiple neural network models in predicting cellular traffic in this study. The fully connected sequential network (FCSN), one-dimensional convolutional neural network (1D-CNN), single-shot learning LSTM (SS-LSTM), and autoregressive LSTM (AR-LSTM) are proposed in the temporal analysis. A 2-dimensional convolutional LSTM (2D-ConvLSTM) model is also proposed in the spatiotemporal framework to forecast cellular traffic over the next 24 hours. The 2D-ConvLSTM model, which can capture spatial relations via convolution operations and temporal dynamics through the LSTM network, is used after creating geographic grids. The results reveal that FCSN and 1D-CNN have comparable performance in univariate temporal analysis. However, 1D-CNN is a smaller network with less number of parameters. One of the other benefits of the proposed 1D-CNN is having less complexity and execution time for predicting traffic. Also, 2D-ConvLSTM outperforms temporal models. The 2D-ConvLSTM model can predict the next 24-hour traffic of internet, sms, and call with root mean square error (RMSE) values of 75.73, 26.60, and 15.02 and mean absolute error (MAE) values of 52.73, 14.42, and 8.98, respectively, which shows better performance compared to the state of the art methods due to capturing variables dependencies. It can be argued that this network has the capability to be utilized in network management and resource allocation in practical applications