基于属性约简和SVM参数优化的入侵检测方法

摘要

SVM is strongly applicable to small sampled, nonlinear or high dimensional classification issues. But there are alsodisadvantages with SVM such as long training period, too much storage occupation by sample sets and so on. The thesis proposes an attribute reduction and parameter optimization based SVM intrusion detection method, which uses the rough set theory to execute feature reduction on sample sets and uses the improved network search algorithm to optimize SVM parameters, so that removes those properties that don't have impact on intrusion detection. Hence the disadvantages like long training period and large storage requirement are overcome. Experiment using KDD99 data set shows that the new method is an effective intrusion detection method. It not only accelerates the training speed, but also improves the accuracy of intrusion detection.%支持向量机( SVM)对于小样本、非线性、高维等分类问题,具有较强的适用性.但是SVM存在训练时间长,样本集占用存储空间过大等问题.提出一种基于属性约简和参数优化的SVM的入侵检测方法.利用粗糙集理论对样本集进行特征约简并使用改进的网格搜索算法对SVM参数进行优化,删除对入侵检测无影响的属性,从而解决SVM训练时间长以及存储空间大的问题.KDD99数据集下的实验表明,该方法是有效的入侵检测方式,不仅加快训练速度,还提高入侵检测的准确率.