Mobile data traffic forecasting in UMTS networks based on SARIMA model: The case of Addis Ababa, Ethiopia
Samuel Medhn, Bethelhem Seifu Shawel, Amel Salem, Dereje Hailemariam · 2017
In planning, operating and developing mobile data networks, one crucial factor is the telecommunication demand that includes number of subscribers and their required service data rates. This demand should be predicted accurately to capture the subscribers' needs to create customer satisfaction. In the Ethiopian context, Ethio Telecom is the sole telecom service provider with over 50 million subscribers for now and is expanding its networks to reach over 100 million subscribers in the next few years. The mobile traffic forecasting approach used in Ethio Telecom, to the best of our knowledge, does not fully consider the data demand already recorded in its core network. This paper presents Seasonal Auto-Regression Integrated Moving Average (SARIMA) model as an alternative way of forecasting Universal Mobile Telecommunication System (UMTS) data traffic taking the city of Addis Ababa, Ethiopia, as a case study. The approach in this paper involves first investigating the past UMTS data traffic load collected from the operator's core network to find an appropriate model that describes the inherent characteristics of the data traffic and secondly, use the model for future prediction. Based on our findings, from the possible candidates SARIMA (2, 0, 1) × (0, 1, 1)7is selected to be the best model with a mean absolute percentage error (MAPE) of 1.17% for 30 days forecast. In addition, the prediction performance comparison of SARIMA (2, 0, 1) × (0, 1, 1)7with respect to SARIMA (2, 0, 1) × (0, 1, 2)7was done resulting in 4.1% of MAPE improvement. These findings will be useful for planning subsequent infrastructure expansions in a way that guarantees better customer satisfaction.