摘要
从特征抽取的角度研究提高入侵检测性能问题,提出应用核偏最小二乘(KPLS)进行入侵特征抽取和检测的方法.其优点在于KPLS能非线性地抽取输入特征的多个正交分量,并保持与输出类别的相关性,可同时完成入侵特征抽取和判别.将该方法应用于基于Linux主机的入侵检测实验,取得了比SVM和KPCR等方法更好的效果.
A novel KPLS based network intrusion feature extraction and detection approach is put forward, among which KPLS serves as simultaneously a non-linear feature extractor and a decision maker. KPLS approach bears the merits that it can not only extract orthogonal score vectors from explanatory variables, but also remain good correlation with response variables. The feature extraction procedure and decision making procedure can be achieved at one time. The novel method is applied to an up-to-date Linux-hosted IDS experimental system and better performance is attained in comparison to SVM and KPCR etc.
出处
《控制与决策》
EI
CSCD
北大核心
2005年第3期251-256,共6页
Control and Decision
基金
国家重点基础研究发展规划项目(2002CB312200)
国家自然科学基金项目(69974014)
教育部高校博士点基金项目(20040251010).
