基于机器学习的Co-training流量分析系统设计与实现  

Design and implementation of Co-training traffic analysis system based on machine learning

作　　者：张远 徐逸卿[2] Zhang Yuan1, Xu Yiqing2(1.Nanjing Institute of Information Technology, Nanjing 210000, China; 2.Nanjing Forestry University, Nanjing 210037, Chin)

机构地区：[1]南京信息技术研究院,江苏南京210000 [2]南京林业大学,江苏南京210037

出　　处：《江苏科技信息》2018年第15期41-44,共4页Jiangsu Science and Technology Information

摘　　要：如何分析流量是一个热门问题,近几年许多国内外的研究学者也对该问题有不少的研究与实践。现有的处理方法大多以基于机器学习的流量识别技术为主,在这些机器学习方法中,SVM技术表现出训练时间短、泛化能力高等优势。但其主要不足在于:需要的样本标记数量多,导致需要花费的成本高。因此文章提出一种基于SVM和Co-training的恶意流量检测方法,该方法引入Co-training半监督方法以降低样本标记数量,同时保持分类的准确性。How to analyze traffic is a hot issue. In recent years, many scholars at home and abroad have also studied and practiced this issue. Most of the existing processing methods are based on machine learning flow recognitiontechnology. Among these machine learning methods, SVM technology has the advantages of short training time and highgeneralization ability. However, its main disadvantage is that it requires a large number of sample tags, resulting in highcosts. Therefore, this paper proposes a malicious traffic detection method based on SVM and Co-training. This method introduces a Co-training semi-supervised method to reduce the number of sample tags while maintaining the accuracy of classification.

关 键 词：流量分析 机器学习 SVM 

分 类 号：U285.49[交通运输工程—交通信息工程及控制]