Network Traffic Forecasting Based on Fixed Telecommunication Data Using Deep Learning
Mousa Alizadeh, Mohammad Taghi Hamidi Beheshti, Amin Ramezani, Hadis Saadatinezhad
Abstract
Network traffic forecasting means estimating future network traffic from previous traffic observations. Network traffic analysis has various applications in a wide range of fields, and considerable research attention has been paid to this area in recent years. Accurate forecasting of network traffic plays an important role in network management and improving the Quality Of Services (QoS). For this purpose, various techniques have been applied such as neural network-based methods and data mining methods. This paper has concentrated on examining various methods of analyzing and forecasting network traffic based on deep learning. Therefore several Recurrent Neural Network (RNN) models such as Random Connectivity Long Short-Term Memory (RCLSTM), Gated Recurrent Unit (GRU), and some Feed Forward Neural Networks (FFNN) like Multi-Layer Perceptron (MLP) and Convolutional Neural Network (CNN) have been studied. Also, the combination of Long Short-Term Memory (LSTM) and MLP as a new method have been proposed. The simulation results have been implemented in Python and compared with other previous algorithms, which shows the high effectiveness and performance of the new approach.