基于成对约束扩展的半监督网络流量特征选择算法

针对网络流量特征选择过程中监督信息缺乏的问题,提出一种基于成对约束扩展的半监督网络流量特征选择算法。该算法同时考虑少量成对约束和大量无标记样本,利用样本集合间的相关性和自相关性,扩展成对约束集到无标记样本上,产生更多可靠性强的成对约束,以揭示样本空间分布信息。最后,利用扩展的成对约束集进行特征选择。实验证明:与未进行成对约束扩展的算法相比,该算法在少量初始成对约束的情况下能获得更好的分类性能。

Semi-supervised network traffic feature selection algorithm based on extension of pairwise constraints

Aiming at problem that lack of supervised information in network traffic feature selection process,a semi-supervised network traffic feature selection algorithm based on extension of pairwise constraints is proposed.Fully considering a small mount of pairwise constraints and a large of unlabeled samples,this algorithm extends pairwise constraints sets to unlabeled samples based on inter-set and intra-set,produces more reliable pairwise constraints to reveal distribution information of the sample space.Compared with algorithms without pairwise constraints extension,experiment shows that the proposed algorithm can achieve better classification performance with fewer initial pairwise constraints.