线性逼近SVM在入侵检测中的应用

Application of Linear Approximation SVM in Intrusion Detection

引入分段线性识别算法,提出一种线性逼近支持向量机(SVM)入侵检测模型。将特征空间剖分成若干子空间,在每个子空间中基于SVM构造5个最优分类面,将各个分类面链接起来构成5个分片最优分类面以逼近理论上的最优分类超曲面。实验结果证明,该模型的训练时间较短,在噪声数据存在的情况下识别正确率较高。

Piecewise linear recognition algorithm is introduced in this paper,and an linear approximaiton Support Vector Machine（SVM） model for intrusion detection is proposed.In this model,the feature space is partitioned into several sub-space,the five best face are made in each sub-sector based on support vector classification,and then link together to form each of five categories face the optimal classification surface patch to approximate the theoretical optimal separating hyper surface.Experimental results show that the model needs short training time,and in the presence of noise data,it has high detection accuracy rate.