基于协同增量支持向量机的网络入侵检测

Network Intrusion Detection Based on Cooperative Incremental Support Vector Machine

针对网络行为数据的不均衡、数量大、更新快的问题,结合均衡化、增量学习、分类器集成思想提出一种用于网络入侵检测的协同增量支持向量机算法,该算法利用多个分类器的协同工作,提高算法速度,每个子分类器依据类样本的空间分布计算类样本错分代价,避免分类超平面偏移,对多个子分类器进行加权集成获得最终分类器,提高最终分类器在不均衡数据集下的分类性能.最后,在KDDCUP1999数据集上的仿真实验结果表明,该算法对整体准确率、少数类及未知攻击都有较高的检测准确率.

This paper focused on the imbalanced dataset,large sample size and quick update of attack method,proposed a network intrusion detection based on cooperative incremental support vector machine. This method constructs each sub-classifier and calculates spatial distribution of samples. It can reduce the offset of the classified hyperplane and increase algorithm speed,obtain the final classifier according to the weight ensemble classifiers.Finally,experiment results on KDDCUP99 dataset show this method has higher detection accuracy for unknown attacks.