基于Re-FCBF的入侵特征选择算法研究  

Research on Feature Selection Algorithms in Intrusion Detection Based on Re-FCBF

作　　者：黄春虎 努尔布力[1] 解男男[2] 胡亮[2] 

机构地区：[1]新疆大学信息科学与工程学院,乌鲁木齐830046 [2]吉林大学计算机科学与技术学院,长春130012

出　　处：《激光杂志》2016年第1期103-107,共5页Laser Journal

基　　金：国家自然科学基金项目(61163052;61303231;61433012);国家自然科学基金联合基金项目(U1435215)

摘　　要：随着网络数据量的日益增大从而导致计算复杂和时间复杂度高的问题,特征选择是降低高维数据计算复杂度的有效方法之一,本文提出一种面向入侵检测领域的两阶段特征选择方法,通过Relief F消除冗余特征和FCBF算法选择特征子集的结合方法,目标是达到在不减少入侵检测效果的情况下通过降维减少时间复杂度。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,结果表明所提出的方法能够在保证分类准确率的情况下,减少至少20%计算时间。With the network data quantity increasing, there are problems about high time complexity and computa- tional complexity, feature selection is one of the effective methods to reduce the computational complexity of the high di- mensional data,Thus, the combining method is which the ReliefF eliminated the redundant features and FCBF selected feature subset, proposed a two stage feature selection method for intrusion detection field, the goal is to reduce the time complexity by dimensionality reduction without reducing the intrusion detection effect. Experiments used KDD CUP 99 dataset ,and selected the data of the 41 dimensional features. Support Vector Machine was adopted as classifier,the re- sults showed the proposed method can reduce at least 20% computational time and ensure the accuracy of classification.

关 键 词：入侵检测 特征选择 RELIEFF FCBF 

分 类 号：TN911[电子电信—通信与信息系统]