关键词: 特征约简；相关性分析；特征重要性评价；主成分分析(PCA)；支持向量机

Abstract:

In an intrusion detection system, a large number of samples with highdimension are analyzed. Reduction of dimensions of the samples is crucial for reducing training time and improving accuracy and realtime capability. In this paper, two feature reduction methods, feature correlation analysis and feature importance measurement, are proposed. Support vector machine is used as a classifier to evaluate effectiveness and performance of different feature reduction methods. Experimental results show that, comparing with the principal component analysis (PCA) method, the two described methods permit realtime processing with high classification accuracy. Moreover, the feature importance measurement in data preprocessing time, training time and classification accuracy is equivalent to that of PCA, and better than the feature correlation approach.

Key words: [WT5HZ]Key words[WT5BZ]： feature reduction; feature correlation analysis; feature importance measurement; principal component analysis (PCA); support vector machine