A Comparative Analysis of Telecommunication Network Traffic Forecasting: A Three Model Approach

The uplink traffic demand is rising due to the recent high demand of internet traffic, upsurge of social media networking and the increase of the smartphone usage. However, most of the existing works on traffic prediction have concentrated on the downlink with minimal attention on the uplink. Secondly, in the literature ensemble models have been found to generate better prediction values in general than the individual models. This study therefore investigates telecommunication network uplink traffic through the application of three individual forecasting models. In this approach, hourly, daily and weekly uplink traffic of 3G mobile networks as measured from an operator were analysed and forecasted using ARIMA, Elman Neural Network (ENN) and Multilayer Perceptron (MLP) models. With hourly 3G uplink traffic ARIMA (0,1,5), ENN (1-25-1) and MLP (1-25-1) models were selected as the most effective model for prediction, ARIMA (4,1,1), ENN (1-24-1) and MLP (1-24-1) were the best for daily traffic while ARIMA (2, 1,2), ENN (1-10-1) and MLP (1-35-1) are the best models for weekly traffic. The model selection was based on the minimum values of prediction accuracy measures. The comparative results indicated that ARIMA, ENN and MLP models can be applied to predict hourly, daily and weekly 3G uplink traffic effectively.