入侵检测系统中基于主成份分析特征提取研究

Study on Feature Extraction Based on Principal Component Analysis in Intrusion Detection System

数据降维是基于模式识别方法的入侵检测系统需要解决的一个问题。由于主成份分析方法具有两个我们期望的特性,一是不同的主成份之间互不相关,二是每个主成份都是所有原始特征的线性组合,所以将主成份分析应用到入侵检测系统的特征提取中。首先我们使用ReliefF算法去除原始特征中与分类无关的特征,然后再进行主成份分析。在实际的数据集KDDCUP’99上进行的实验结果表明提出的方法是有效及实用的。

The first problem which should to be solved in intrusion detection system based on pattern recognition method is reducing data dimentions. Because principal component analysis has two attributes we expected, one is that various principal component is not relevant, and the other is that each principal component is linear combination of all original features, the authors apply the principal component analysis to extracting features from intrusion detection system. Firstly, we use ReliefF to get rid of features that are irrelevant with classification from original features, then employ principal component analysis. The experimental results on real KDD CUP＇99 dataset show that the proposed method is effective and practicable.