面向移动互联网流量的行为特征和自适应分类方法研究

项目名称： 面向移动互联网流量的行为特征和自适应分类方法研究

项目编号： No.61501128

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 无线电电子学、电信技术

项目作者： 刘珍

作者单位： 广东药科大学

项目金额： 19万元

中文摘要： 当前移动互联网流量呈爆炸式增长，面临如何保障终端应用的服务质量和安全等问题。互联网流量分类技术是实施网络管理的重要基础，但是目前大多研究仍然聚焦于有线网的流量分类技术，它们难以有效运用于应对移动互联网面临的问题，困难主要表现在：1）传统L7-filter方法难以胜任新型终端应用流量的类别标记，移动网络流量类别标记困难；2）移动网络环境易变，流量统计特征也随之变化，流量分类器难以适应网络环境的变化。本项以突破上述瓶颈为目标，研究适合移动终端应用的流量类别归类规则，设计移动互联网流量的类别标记方法，进而构建基准数据集；研究移动终端应用的行为建模方法，在行为模型上提取较稳定的行为特征；采用行为特征描述基准数据集，构建流样本集；在此基础上，利用集成学习、负相关学习和概念漂移探测理论，研究能自适应网络环境的流量分类方法。通过本项研究，为移动互联网流量分类技术的研究提供理论基础和科学依据。

中文关键词： 移动互联网流量分类；行为特征；集成学习；概念漂移；多类不平衡

英文摘要： As mobile network traffic becomes larger and larger nowadays, mobile network faces how to guarantee the QoS and security of terminal apps. Internet traffic classification techniques have been shown as the important foundation of network management. However, most existing traffic classification techniques are based on wired network, which cannot be directly used for handling the challenges faced by mobile network. The obstacles are as follows. 1) It is hard to labeling the mobile network traffic class, as L7-filter cannot effectively build the ground truth for the traffic generated by new terminal apps. 2) Most flow statistical features are not stable as the mobile network environment changes constantly, and the traffic classifier cannot adapt the change of network environment. By targeting on breaking through these bottlenecks, we study the taxonomy rules of traffic classes and devise a method for obtaining the ground truth of mobile network traffic, so as to build the benchmark datasets. Through researching on the behavior modeling method for mobile terminal apps, we extract stable behavior features from behavior model. The behavior features are used to characterize the benchmark datasets and further build flow sample sets so as to be used for the following research. Applying ensemble learning, negative correlation learning and concept drift detecting theories, we study the adaptive traffic classification method. This project will provide basic theories to guide future researches on mobile network traffic classification techniques.

英文关键词： mobile network traffic classification;behavior features;ensemble learning;concept drift;multi class imbalance