论文部分内容阅读
流分类技术在网络安全监控,QoS,入侵检测等方面起着重要的作用。流分类器处理的数据含有大量的相关与冗余特征,这不仅增加了分类器的计算复杂性,同时也影响了分类器的分类效果。针对高维特征空间,特征选择一方面可以提高分类精度与效率,另一方面可以找出富含信息的特征子集。该文提出一种wrapper型特征选择算法VFSA-C4.5来构建轻量级的流分类器。该算法采用快速模拟退火VFSA搜索策略对特征子集空间进行随机搜索,然后以提供的数据在C4.5上的分类正确率作为特征子集的评价标准,来获取最优特征子集。在流数据集上进行的大量实验结果表明,基于VFSA-C4.5的流分类器在不影响分类性能的情况下能够提高分类速度。
Traffic classification technology plays an important role in network security monitoring, QoS and intrusion detection. The data processed by the stream classifier contains a large number of correlation and redundancy features, which not only increases the computational complexity of the classifier, but also affects the classification effect of the classifier. For high-dimensional feature space, feature selection can improve classification accuracy and efficiency on the one hand, and find a subset of feature-rich information on the other hand. This paper proposes a wrapper-based feature selection algorithm VFSA-C4.5 to construct a lightweight stream classifier. The algorithm uses fast simulated annealing (VFSA) search strategy to search the feature subset space randomly, and then obtains the optimal feature subset based on the classification accuracy of the data provided on C4.5 as the evaluation criteria of the feature subset. A large number of experimental results on stream datasets show that the VFSA-C4.5-based stream classifier can improve the classification speed without affecting the classification performance.