Multi Factor Based Mobile Voice and Data Traffic Forecasting Using Artificial Neural Networks

Abstract

As traffic levels in a mobile network increase, it is important for the Mobile Network Operators to adequately forecast future traffic demands. Strategic planning by the service provider, will result in timely rollout the required capacity for adequate future service provision which will alleviate network congestion and ultimate failure of the systems in place. This will ensure that acceptable Quality of Experience (QoE) and Quality of Service (QoS) is achieved by the deployment and rollout of systems and equipment and optimization of radio access resources in a timely manner. In this research, artificial neural networks were used to forecast mobile traffic demands. The artificial neural network (ANN) was trained on input data consisting of a combination of quantified values of factors influencing mobile voice and data traffic together with the related target data variables. The taught artificial neural network was subsequently analysed on its ability to make forecasts based on the relationships obtained during training of the network. The ANN was used to predict the network busy hour for voice and data traffic per Long Term Evolution (LTE) relay nodes, Radio Network Controller (RNC) and Base Station Controller (BSC) and which could be used to determine the maximum demand for network resources that the network would be subjected to and how robust the system was to traffic growth and the capacity designs that need to be implemented to ensure continuity of service within QoS and QoE limits. The data was sourced from one local telecommunication service operator and the research considered various network nodes (BSC, RNC, LTE relay nodes) and cellular voice and data service technologies. Data from 21 BSC, 10 RNC and 1 LTE nodes was used. The research involves the use of multiple factors as inputs to the neural networks that can be used to establish additional non-linear relationships to the associated targets and therefore improve the accuracy of forecasts. The results of the performance of the trained networks had average MSE values that ranged from 0.950694 to 0.982268 for cellular traffic forecasts (voice and data) and for prediction of the bouncing busy hour (BBH), the range spread from 0.049477 for LTE Data BBH and a maximum of 0.67827 for RNC voice (BBH) which demonstrate the accuracy and applicability of the use of ANN in the forecasting of cellular traffic and associated busy hours.