Mobile Internet Traffic Analysis And Prediction Based On PU Learning

In recent years,with the development of mobile communication technology and the popularization of mobile terminals,the number of mobile users has surged.Millions of mobile users generate massive mobile traffic records every day.How to analyze such a large amount of data efficiently and accurately so as to fully understand the characteristics of mobile Internet traffic,master the Internet behavior of mobile users,and provide users with more high-quality mobile network services has become a concerned research topic for mobile operators and Internet content providers.This paper studies the mass mobile Internet traffic data analysis and prediction through a combination of distributed big data processing technology and machine learning methods.Mainly focusing on two of these issues,one is the prediction of the interaction between mobile users and servers and the other is the identification of mobile video traffic.Through the analysis,we found that both of the two problems have the same characteristics:they can be treated as binary classification problems,and the data are composed by a small number of positive examples and a large number of unlabeled examples.Therefore,this paper try to solve it by using the PU learning algorithm which is often used in the text classification problem.The main innovations of this article are as follows:(1)For mobile user and server interaction prediction,we designed a 194-dimensional feature vector to characterize each user and server connection record,and proposed an improved Spy-based PU learning algorithm.On the Spark platform,a K-means based PU learning algorithm,a Biased-SVM based PU learning algorithm,and an improved Spy-based PU learning algorithm are designed and implemented based on the MLlib machine learning library.Experiments on real mobile traffic data demonstrate that the PU learning algorithm can predict the user and server interaction with good performance.(2)For mobile video traffic identification,we design a 105-dimensional feature vector to represent each mobile traffic record and filter some mobile video-related data from the real mobile traffic data as the experimental data set of this task.The performance of three PU learning algorithms designed and implemented on the Spark platform is analyzed experimentally,and the accuracy of mobile video traffic identification based on PU learning algorithm is verified.This is of great help for analyzing mobile Internet traffic more efficiently and precisely,thus improve network service quality.