Deep-Learning Based Mobile-Traffic Forecasting for Resource Utilization in 5G Network Slicing

Abstract

Network slicing is the key technology in 5G wireless communication, which aims to provide services based on latency, availability, reliability, throughput and more. With the rapid development of mobile networks and new networking applications, it is turning out to be more difficult to meet the Quality of Services (QoS) under the current mobile traffic and mobile-network architecture. Mobile Traffic forecasting is one of the domains that can benefit the mobile companies in optimizing their resources. In this paper, we consider a dataset with Internet usage patterns by users over a period of six days. Based on past time-steps trends we tries to predict the current network slice that would be classified into streaming, messaging, searching, and cloud classes. We compared the four deep learning models namely MLP, Attention-based Encoder Decoder, GRU and LSTM and we evaluated these models on recall, precision and f1 score performance matrices. We found that MLP, Encoder-Decoder models performed average for mobile-traffic forecasting while the GRU, LSTM performs well and out of them LSTM obtained best result.