基于数据引力分类方法的互联网非平衡流量早期识别研究

项目名称： 基于数据引力分类方法的互联网非平衡流量早期识别研究

项目编号： No.61472164

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 自动化技术、计算机技术

项目作者： 彭立志

作者单位： 济南大学

项目金额： 80万元

中文摘要： 互联网中各类流量天然地呈不平衡分布状态，在非平衡流量发生的早期阶段，对其进行准确高效的识别对网络管理与网络安全都有极其重要的意义。传统识别方法识别非平衡流量时，具有漏报率高等重大缺陷。因而，针对互联网非平衡分布流量，研究有效的识别方法准确高效地识别非平衡流量，是流量识别与应用中的一个亟待解决的问题。本项目就这一问题，从基础理论与应用技术两条主线上系统地开展研究：（1）理论上，基于数据引力分类方法，研究高效的非平衡数据引力分类方法，寻求非平衡流量的有效识别算法；（2）技术上，首先针对具有准确背景信息的流量数据采集困难问题，研究流量样本的背景信息标记技术；基于有效的基础数据，研究非平衡流量早期特征的提取方法与有效性验证；进而通过有效特征数据，研究单目标类型与多目标类型非平衡流量识别技术；最后基于这些关键技术，研发具有实际应用价值的非平衡流量识别系统，为互联网非平衡流量识别提供系统的解决方案。

中文关键词： 流量识别；非平衡分类；数据引力分类；机器学习；网络测量

英文摘要： Different type of traffics in Internet show imbalanced distributions naturally. It is very important for network management and security to identify imbalanced Internet traffics effectively at their early stage. Traditional identifying techniques suffer from high false negative rates when applying for imbalanced traffics. Thus, it is a critical problem to be resolved for Internet traffic identification research to study early stage imbalanced traffic identification systematically. In this project, we will study the early stage imbalanced traffic identification problem theoretically and technically: (1) For theoretical aspect, we are set to seek an effective imbalanced model for the data gravitation classification model; (2) For the technical aspect, we will firstly study the accurate ground information labeling technique of Internet traffic data. Based on the accurate traffic data, we will study the early stage feature extraction method for imbalanced traffics and the validations of the feature sets. And then, by using the feature data, we will study imbalanced identification techniques for single-type and multi-type of Internet traffics. At last, based on the aforementioned techniques, we will develop an applicable prototype of imbalanced traffic identification system. Our works could provide a systemic solution for the imbalanced Internet traffic identification problem.

英文关键词： Traffic Identification;Imbalanced Classification;Data Gravitation Classification;Machine Learning;Network Measurement