User Mobility Pattern Prediction Based On Massive Mobile Traffic Data

The Mobile Internet in recent years has been developing rapidly and brings exponential growth of mobile traffic data which has significant features of big data.Using mobile traffic data to analyze the behavior of mobile users and improving the quality of network service has become a hot topic in the area of mobile Internet research.Various kinds of big data technologies have emerged in an endless stream and brought a great convenience for the transmission and processing of large data.Among vavious kinds of tools,Hadoop is an effective and popular platform to conduct large-scale data processing,and Flume is a log aggregation system providing distributed,reliable and available service.The thesis firstly introduce the large-scale data processing tools Hadoop and the two basic components of Hadoop:HDFS and MapReduce programming framework.Then Flume is also introduced as an available service for moving large amounts of streaming event data.The large amounts of data needs a distributing system to transfer from collecting module to Hadoop cluster.In this thesis,we design two didstributing solutions which are based on Apache Flume and self developed Importer respectively.We also compare their advantages and disadvantages.The analysis of user behavior based on massive traffic data is significant for Internet service providers(ISP).Among different kinds of traffic analysis,understanding user mobility pattern and predicting their locations are of great importance in many areas.By extracting user mobility pattern,network operators can provide efficient network planning for mobile phone users and then develop more reliable communication protocols.In this thesis,with real traffic data collected from a service provider in southern China,we have a research on user mobility pattern from both group and individual perspectives.In the analysis of user group mobility,we consider all users as a whole unit.With real mobile network traffic data,we analyze the amount of user access,roaming capacity,hot cells and etc.In the analysis of user individual mobility,we focus on predicting user mobility patterns based on their different mobility characteristics.We present the Intelligent Time Division(ITD)method and Time-Based Markov(TBM)predictor for the location prediction of the stationary and mobile users respectively.With three-consecutive-week data collected from Long Term Evolution(LTE)mobile network,we cluster users into stationary one or mobile one with an entropy-based method for distinguishing groups with distinct mobility characteristics,and then we use ITD method and TBM predictor on two kinds of users respectively.Experiments demonstrate the effectiveness and better performance of our proposed methods compared with the baselines,as well as the adaptabilities of different predictors according to individual's mobility characteristics,which is of great importance for ISPs or Location Based Service Providers.