Traffic Data Analysis And Forecasting In Mobile Communication Networks

Nowadays, mobile communication technology is highly developed and ICT (In-formation Communications Technology) industry gradually grows mature. This leads to the explosive growth of mobile communication market, which mainly consists of smart phones and applications. It is believed that the traffic load of mobile communi-cation networks will grow at a fairly high speed. Thus mobile communication network faces great challenge to meet the QoS (Quality of Service) demand of users. Recently, data mining techniques have achieved a number of successful cases, revealing the great value in data. Big Data Analytics emerges and attracts focus from all sides of societies. The massive data produced in mobile communication networks of different types and in different forms can be an opportunity to researchers and operators. Properly analyz-ing big data in mobile communication networks can contribute to design of network architecture, improvement of protocol, optimization of operation and etc. Thus big data analytics in mobile communication networks becomes an emerging research field and also a hot topic.On the background mentioned afore, mobile communication network data analysis and forecasting is investigated in this paper based on a real dataset. The work consists of two parts:data analysis of large scale mobile communication network and load pre-diction of base stations.Multiple data analysis techniques and tools are utilized to discover the characteris-tics of mobile communication networks from different aspects. Cumulative distribution function is used to analyze the load distribution of the whole network and individual base station respectively. The imbalance of load is revealed and the distributions of different base stations vary. In temporal characteristic analysis, non-stationarity and inherent pe-riodicity of most base stations is verified with time series autocorrelation. By Moran’s I calculation, global spatial autocorrelation is studied, which shows the daily repeat of base station load evolution in space. Local spatial autocorrelation with LISA cluster map reveals the non-stationarity of traffic load in local area. In the end different types of base stations are classified using cluster analysis of base stations. Combined with location mapping of different types of base stations, a unique community is identified.The most direct application of mobile network data so far is to forecast future load based on measurement. A base station traffic forecasting scheme incorporating infor-mation of neighbors is proposed in this paper. In the proposed scheme, support vector regression is chosen as the basic model. The parameters of the prediction model are de-cided by the Partial Swarm Optimization algorithm. To incorporate the information of neighbors, grid search is utilized to determine the neighboring base station set and his-torical length is determined through testing. The proposed forecasting scheme is tested on the real dataset. Results show that the proposed scheme can achieve better prediction accuracy. In the end, the application of the proposed forecasting scheme is discussed. The proposed forecasting method can be used in the energy saving plans based on traffic prediction in green communications.Conclusions drawn from base station load analysis can promote the understanding of mobile communication networks and analysis method used can also be utilized in other analysis works of mobile network data. The proposed base station load prediction scheme is meaningful to study of base station energy saving plans.