入侵检测系统中基于变量相似性特征选择

ReliefF是一种在很多场合经常使用的filter式的特征选择方法．然而该方法的一大缺点是不能辨别冗余特征。基于ReliefF算法提出一种混合的有监督的特征选择算法。该算法首先利用ReliefF算法去除与分类无关的以及权重低于一定阈值的特征，然后采用一种变量相似性准则来去除冗余特征。在实际的数据集KDDCUP＇99上进行的实验结果表明该混合特征选择方法较单独使用ReliefF方法在分类精度上有一定的提高。

Feature Selection Method Based on Variable Similarity in Intrusion Detection System

ReliefF is a filter's feature selection method which is often used in many cases, while the method cannot discriminate redundant features which is a serious defect. This paper proposes a mixed and supervised feature selection algorithm based on ReliefF. The algorithm first uses ReliefF to get rid of irrelevant features and the features which have lesser weight value than predetermined threshold, then uses a variable similarity measure for eliminating redundant features. The experimental results on real KDD CUP'99 dataset show that this mixed feature selection method is superior to the method using ReliefF solely on classification accuracy.