一种快速KMSE算法及其在异常入侵检测中的应用

范自柱^1,2, 徐勇¹, 徐保根², 朱旗¹

1.哈尔滨工业大学深圳研究生院;2.华东交通大学基础科学学院



摘要:

为提高核最小均方误差(Kernel Minimum Squared Error,KMSE)方法的计算效率,利用特征空间中不相关的样本矢量("基样本"),提出了一种快速KMSE算法,并利用"基样本"与一个样本间的核函数对该样本抽取特征.在入侵检测数据集KDDCUP1999和其他基准数据集上实验表明:该方法不仅高效,并且分类和检测效果良好,"基样本"只占训练样本的很小一部分比例,使用它们可以显著提高特征抽取效率.

关键词:  核最小均方误差  鉴别分析  快速算法  线性相关  入侵检测

DOI：10.11918/j.issn.0367-6234.2011.03.019

分类号:TP393.08

基金项目:国家自然科学基金资助项目(61071179,11061014,60803090,61001037,61065003);教育部新世纪优秀人才支持计划资助项目(NCET-08-0156,NCET-08-0155);江西省自然科学基金资助项目(2010GQS0027)



A fast KMSE algorithm and its application on anomaly intrusion detection

FAN Zi-zhu^1,2, XU Yong¹, XU Bao-gen², ZHU Qi¹

1.Shenzhen Graduate School,Harbin Institute of Technology,518055 Shenzhen,China;2.School of Basic Science,East China Jiaotong University,330013 Nanchang,China)

Abstract:

To improve the computational efficiency of the kernel minimum squared error(KMSE) algorithm,we propose a fast KMSE by using the uncorrelated sample vectors,known as basic samples,in the feature space.And we describe the theoretical relationship between the linear correlation of sample vectors in the feature space and the determinant of the kernel matrix.The kernel functions between a sample and the basic samples were used to extract the features from the sample.The whole basic samples is only a small portion of the training set.The experiments on the intrusion detection dataset KDDCUP1999 and other benchmark datasets show that the fast KMSE is computationally efficient and can achieve high classification accuracy.

Key words:  Kernel Minimum Squared Error(KMSE)  discriminant analysis  fast algorithm  linear correlation  intrusion detection