一种新型混合特征选择方法及其在入侵检测中的应用  

A New Hybrid Attribute Selection Method and Its Application in Intrusion Detection

作　　者：毛俐旻[1] 姚淑萍[1] 胡昌振[1] 

机构地区：[1]北京理工大学计算机网络攻防对抗技术实验室,北京100081

出　　处：《北京理工大学学报》2008年第3期218-221,共4页Transactions of Beijing Institute of Technology

基　　金：国家部委基金资助项目(20021823)

摘　　要：针对高维数据包含的不相关和冗余特征影响检测方法性能的问题,提出了集成filter和wrapper方法的混合特征选择新方法.采用基于信息增益的filter方法,删除不相关特征;采用基于改进的自适应遗传算法和评价函数的wrapper方法,获取最优特征子集.在入侵检测中的应用表明,该方法能降低特征选择的时间,检测率和虚警率均优于其它方法.The performance containing irrelevant and of a detection method is degraded by the high dimensional data redundant attributes. A new hybrid attribute selection method integrating filter and wrapper methods is proposed. Filter method based on mutual information is firstly used to remove irrelevant attributes. Wrapper method based on improved adaptive genetic algorithm and improved evaluation function is used to select optimal attribute subset. Applications in intrusion detection showed that this approach can reduce the time of attribute selection and has better performance in terms of true positive rate and false positive rate than other methods.

关 键 词：特征选择 filter方法 wrapper方法 入侵检测 

分 类 号：TP393.08[自动化与计算机技术—计算机应用技术]