摘要
引入类均值向量度量及αβ指数分布方法,旨在提高分类正确率的基础上,克服由于抽样而带来的对分类结果的影响。利用流记录NOC_SET为DATASET,并以NETFLOW固有的测度和少量扩展测度为属性,利用所提出的FBRI(Flow behavior identification)属性选择算法对经典的机器学习算法进行流量识别。实验结果表明:任意比例的抽样对于采用FBRI属性选择的评估结果基本一致,并且利用FBRI属性选择算法可以提高应用识别正确率。
The introduction of class mean vector measurements and α β exponential distribution methods is to base on improving the correct classification ratio to overcome the classification result influence brought by sampling. Taking flow records NOC_SET as dataset, in addition taking NETFLOW＇s inherent measure and a few extended measures as features, the method uses the proposed FBRI （ Flow behavior identification） attribute selection algorithm to identify the traffic for classic machine learning algorithms. Experimental results show that the evaluation results to FBRI feature selections for arbitrary proportions of sampling are of the same ; additionally, by using FBRI feature selection algorithm proposed in the thesis, the correct identification ratio can be improved.
出处
《计算机应用与软件》
CSCD
2011年第11期249-253,共5页
Computer Applications and Software
