针对入侵检测领域计算复杂、时间复杂度高的难题,本文通过添加入侵特征间的相关度量准则,进一步改进Re-ReliefF入侵特征选择方法,增强区分特征间互信息的能力,达到更有效降维。目标是在改进ReReliefF算法的基础上,实现更佳的入侵检测效果。实验采用KDD CUP 99数据集,对数据的41维特征进行选择,采用支持向量机作为分类器,结果表明所提出的改进方法在分类的耗时和准确率略好的情况下,提高27%的误报率。 In order to solve the problem of computational complexity and high time complexity in intrusion detection, this dissertation further improves the capability of Re-ReliefF intrusion signature selection by adding correlation metrics between intrusion features to enhance the capability of distinguishing information from each other, thus achieving more efficient dimensionality reduction. The goal is to improve the ReReliefF algorithm based on better intrusion detection results. The KDD CUP 99 dataset was used to select the 41-dimensional feature of the data. The SVM was used as the classifier. The results show that the proposed method improves the time-consuming and accuracy of classification by 27% False alarm rate.