基于近邻传播学习的半监督流量分类方法

准确的流量分类是进行网络管理、安全检测以及应用趋势分析的基础.针对完全监督和无监督分类的缺陷, 提出了一种基于近邻传播学习的半监督流量分类方法.通过引入近邻传播聚类机制构建分类模型, 使得分类器实现过程简单、运行高效. 应用半监督学习的思想, 抽象出少量已标记样本流约束和流形空间先验信息, 定义了流形相似度的距离测度, 既降低了标记流量样本的复杂度, 又提高了流量分类器的性能.理论分析和实验结果表明:算法具有较高的分类准确性和较好的凝聚性.

Semi-supervised Traffic Identification Based on Affinity Propagation

Accurate traffic identification is the keystone of network management, security diagnosis and application prediction analysis. Aiming at the deficiencies of supervised and unsupervised classified methods, we present a novel scheme called semi-supervised internet traffic identification based on affinity propagation (AP). In order to circumvent the problem of choosing initial points, the method introduces affinity propagation clustering to construct classification model simply and effectively. Based on the idea of semi-supervised learning, a few restrictions of labelled flows and priori manifold distribution of sampled space are abstracted. Also, manifold similarity is defined. Henceforth, the semi-supervised method can not only largely reduce the complexity of marking sampled flows, but also nicely improve the performance of the classifier. Theoretical analysis and experimental results show that the algorithm can achieve higher accuracy and better aggregation.