基于对支持向量机的多类分类算法在入侵检测中的应用

发布时间：2018-04-13 08:06

本文选题：对支持向量机 + 多类分类　；参考：《计算机应用》2013年02期

【摘要】：针对基于传统支持向量机(SVM)的多类分类算法在处理大规模数据时训练速度上存在的弱势,提出了一种基于对支持向量机(TWSVM)的多类分类算法。算法结合二叉树SVM(BT-SVM)多类分类思想,通过在二叉树节点处构造基于TWSVM的分类器来达到分类目的。为减少二叉树SVM的误差累积,算法分类前首先通过聚类算法得到各类的聚类中心,通过比较各聚类中心之间的距离来衡量样本的差异以决定二叉树节点处类别的分离顺序,最后将算法用于网络入侵检测。实验结果表明,所提算法不仅保持了较高的检测精度,在训练速度上还表现出一定优势,尤其在处理稍大规模数据时,这种优势更为明显,是传统二叉树SVM多类分类算法训练速度的近两倍,为入侵检测领域大规模数据处理提供了有效参考价值。
[Abstract]:Aiming at the disadvantage of multi-class classification algorithm based on traditional support vector machine (SVM) in processing large scale data, a multi-class classification algorithm based on support vector machine (SVM) is proposed.The algorithm combines the idea of binary tree SVMU BT-SVM) and constructs a classifier based on TWSVM at the node of binary tree to achieve the purpose of classification.In order to reduce the error accumulation of binary tree SVM, the clustering centers are first obtained by clustering algorithm before classification. By comparing the distance between the clustering centers, the difference of samples is measured to determine the classification order of binary tree nodes.Finally, the algorithm is applied to network intrusion detection.The experimental results show that the proposed algorithm not only maintains high detection accuracy, but also shows some advantages in training speed, especially in the processing of slightly large scale data.It is nearly twice the training speed of traditional binary tree SVM multi-class classification algorithm, and provides an effective reference value for large-scale data processing in intrusion detection field.
【作者单位】：南京航空航天大学计算机科学与技术学院;
【分类号】：TP393.08

【参考文献】