适于P2P网络流量识别的SVM快速增量学习方法

A Fast Incremental Learning Method of SVM for P2P Network Traffic Identification

针对标准支持向量机在P2P网络流量识别中不支持增量学习的问题,提出一种适于P2P网络流量识别的SVM快速增量学习方法。在对违背Karush-Kuhn-Tucker条件的新增正负样本集分别进行聚类分析基础上,运用聚类簇中心对支持向量机训练生成一个接近增量学习最优分类超平面的过渡超平面,并以此超平面为基准确定初始训练样本集上非支持向量和支持向量的互相转化,进而生成新的样本集实现SVM增量学习。理论分析和实验结果表明,该方法能有效简化增量学习的训练样本集,在不降低P2P网络流量识别精度的前提下,明显缩短SVM的增量学习时间和识别时间。

In P2P network traffic identification, aims to such the problems that SVM does not support incremental learning. Proposes a fast incremental learning method of SVM for P2P network traffic identification. After clustering of positive and negative training samples that violate Karnsh-Kuhn-Tucker conditions, a temporary classification hyperplane close to classification hyperplane of incremental learning is obtained by using clustering centers to train standard SVM. Based on it, transfers support-vector and non-support-vector in original training samples to produce new training samples for incremental learning of SVM. Analysis and simulation shows that the method effectively simplifies training samples of incremental learning and greatly reduces the training and traffic identification time of SVM in incremental learning.