Abstract: To eliminate the interaction in subsets of intrusion detection data and make the feature fusion of every subset optimized，a feature extraction method based on Classified Principal Component Analysis（CPCA） is presented.Sample data employed is divided into several subsets according to the types of the present attacks，and PCA is performed on every subset separately.The experimental results demonstrate that this method has the merit of fast learning for classifier and higher detection speed.

Key words: intrusion detection, data preprocessing, feature extraction, principal component analysis, feature fusion

摘要： 数据降维是提高入侵检测分类器的学习效率和检测速度的重要手段。针对目前入侵检测数据特征降维力度不够，提出了一种基于主成分分析的分类特征降维方法。该方法把样本集按数据类型分割成多个子集，分别对每个子集进行主成分分析来消除各子集间在降维时的相互影响，使得每个子集的降维达到最佳。实验结果表明采用分类主成分分析方法能够更有效地降低数据维数，提高了入侵检测分类器的学习速度和检测速度。

关键词: 入侵检测, 数据预处理, 特征提取, 主成分分析, 特征降维