高维混合入侵检测数据的单类分类算法研究  

RESEARCH ON ONE-CLASS CLASSIFICATION ALGORITHM FOR MIXTURE AND HIGH DIMENSIONAL INTRUSION DETECTION DATA

作　　者：莫少慧 古兰拜尔·吐尔洪 买日旦·吾守尔[1] Mo Shaohui;Gulanbaier Tuerhong;Mairidan Wushouer(College of Information Science and Engineering,Xinjiang University,Urumqi 830046,Xinjiang,China)

机构地区：[1]新疆大学信息科学与工程学院,新疆乌鲁木齐830046

出　　处：《计算机应用与软件》2022年第4期300-310,共11页Computer Applications and Software

基　　金：教育厅高校科研青年项目(61021800032,61021211418);自治区高层次创新人才项目(100400016,042419006);新疆大学博士启动基金项目(620312308,620312310)。

摘　　要：针对传统的方法未能有效地处理具有高维、混合以及不平衡的特性的入侵检测数据,结合单类分类概念和均值计算方法,提出一种单类分类K近邻(KNN)均值算法。其使用主成分分析混合(PCA mix)方法对高维混合数据进行降维预处理,并运用Bootstrap方法计算决策边界,提高准确性。通过使用三个不同的公开数据集进行实验,并与单类分类支持向量机(OCSVM)算法作对比,其分辨精度提高至94%以上,充分验证了提出算法的优越性。The traditional methods can’t effectively deal with intrusion detection data with high-dimensional, mixture and unbalance characteristics. This paper proposes a k-nearest neighbor(KNN) algorithm by incorporating the concept of one-class classification and average distance calculation. The approach employed the principal component analysis mix(PCA mix) as a preprocessing to reduce the dimension of mixture and high dimensional data. The Bootstrap method was used to calculate the decision boundary for improving the accuracy. A comparison experiment was conducted against the one-class classification support vector machine algorithm using three different public datasets. The results show that the proposed approach can improve the detection accuracy to 94% and higher, which proves the advantage of the proposal.

关 键 词：高维混合 不平衡 入侵检测数据 单类分类 K近邻 PCA mix BOOTSTRAP 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]