网络流量识别中特征选择算法的研究与应用

发布时间：2018-05-29 14:36

本文选题：流量识别 + 机器学习　；参考：《西安电子科技大学》2014年硕士论文

【摘要】：网络应用爆发式增长，网络流量急速膨胀，大量涌现的新型应用比传统应用具有更复杂的结构和流量模式基于流量识别技术，能够细粒度的管理和优化网络，引起了广泛的关注其中，基于流量特征采用机器学习的流量识别技术，具有较高的准确率，成为了近年来流量识别领域的研究热点特征选择通过去除无关冗余的特征，获得最优的特征子集，基于该特征子集能够降低学习算法的复杂度，提升分类的准确率及速度本文首先介绍了流量识别技术机器学习技术及特征选择算法的相关概念，并简单介绍了使用互信息进行度量及SU算法，在此之上提出了两种新的基于互信息的特征选择法： 1.基于互信息的Filter式特征选择法运用改进的SU算法去掉不相关的特征，并基于互信息去掉冗余特征，通过反复调整阈值进行迭代，以提高分类准确率 2.基于互信息的Wrapper式特征选择法运用改进的SU算法去掉不相关的特征，并基于互信息去掉冗余特征，，直接使用分类器的分类准确率作为判断标准来指导算法进行迭代，以获得最佳阈值从而达到最好的分类效果在UCI数据集上的实验结果显示出，本文给出的两种特征选择算法具备较好的分类性能将本文所提出的特征选择法应用于网络流量的类别识别中，在Andrew W.Moore数据集上的实验结果表明，算法在保证了分类准确率的同时，取得了显著的特征约减效果本文选出的流量识别的最优特征子集，能够保证较高的分类性能并大大缩短分类器的分类时间，因此为合理且有效的特征子集
[Abstract]:The network application explodes, the network traffic expands rapidly, a large number of new applications have more complex structure and traffic pattern than the traditional application, which can manage and optimize the network with fine granularity. Among them, traffic recognition technology based on machine learning, which has high accuracy rate, has become the research hotspot in the field of traffic identification in recent years. Feature selection can reduce the complexity of learning algorithm and improve the accuracy and speed of classification by removing redundant features and obtaining the optimal feature subset. In this paper, we first introduce the related concepts of machine learning and feature selection algorithm of traffic recognition technology, and simply introduce the use of mutual information to measure and Su algorithm, and then propose two new feature selection methods based on mutual information. 1. The Filter feature selection method based on mutual information uses the improved Su algorithm to remove irrelevant features, and based on mutual information to remove redundant features, iterates by adjusting the threshold value repeatedly, in order to improve the classification accuracy. 2. The Wrapper feature selection method based on mutual information uses the improved Su algorithm to remove irrelevant features, and based on mutual information to remove redundant features, the classification accuracy of classifier is directly used as the judgement standard to guide the algorithm to iterate. To get the best threshold to achieve the best classification effect. The experimental results on the UCI dataset show that the two feature selection algorithms presented in this paper have good classification performance. The proposed feature selection method is applied to the classification of network traffic. The experimental results on the Andrew W.Moore dataset show that the algorithm not only ensures the classification accuracy, but also achieves the significant feature reduction effect of the optimal feature subset selected in this paper. It can guarantee high classification performance and greatly shorten the classifier's classification time, so it is a reasonable and effective feature subset.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.06;TP181

【共引文献】