Internet Activity Forecasting Over 5G Billing Data Using Deep Learning Techniques

Abstract

Understanding the flexibility of traffic requirements on wireless networks is challenging due to the high density of mobile devices connected to the network. This has made things more difficult given the wide range of devices available and the different types of services they can provide. Internet activity data of 5G billing traffic is an important way to analyze load in a confined area, providing solutions for various 5G infrastructure and applications. In previous decades, deep learning techniques have played a vital role in analyzing such data, and their result consolidates the proof of its veteran performance. The open-source dataset used in this experimentation work is well known as Big data Challenge 2014, which was made publicly available by Telecom of Italia. We evaluate our work with four different networks GRU, LSTM, Bi-directional LSTM and encoder decoder LSTM in which we achieve the lowest mean absolute error in the encoder-decoder CNN-LSTM model with a training loss of 0.0108 and validation loss of 0.0064.