摘要
为了提高支持向量机的托攻击检测效果,提出一种不同于单一算法的基于特征选择和支持向量机的托攻击检测方法。首先定义特征的样本差异性度量,并由此推导出特征的类别可分性度量作为特征选择准则,然后用支持向量机评估所选特征子集的有效性,在不损失样本信息的前提下,通过递归反向特征剔除算法实现检测特征的自动优选,最后利用支持向量机来检测攻击用户概貌。在标杆数据集上与文献中的经典方法进行实验比较和分析,结果显示该方法可以有效地提取最具检测能力的特征子集,同时能获得更好的检测效果,证明了方法的有效性。
In order to improve SVM＇ s shilling attack detection result, we propose a shilling attack detection method, which differs from the single algorithm and is based on feature selection and SVM. First, we define the sample diversity metric of features, and derive from it the classification separability metric of features and use as the feature selection criterion; Then we assess the effectiveness of the selected feature subset with SVM, and implement the automatic opfimised selection of detection features through rccursive reverse features culling algorithm on the premise of no sample information loss ; Finally we use SVM to detect the attacking users profile. Experimental comparison with classic methods in literature and its analysis are done on benchmark dataset, results show that the proposed method is able to effectively extract the feature subset with utmost detection capability, and can achieve better detection result simultaneously, this proves the effectiveness of the method.
出处
《计算机应用与软件》
CSCD
2015年第5期270-272,277,共4页
Computer Applications and Software
基金
中央财政支持地方高校发展专项资金科研项目(DUFE2014Q34
DUFE2014Q36)
辽宁省教育厅科学技术研究项目(L2013435)
教育部人文社会科学研究项目(14YJC63036)
关键词
推荐系统
托攻击检测
特征选择
支持向量机
Recommending system Shilling attack detection Feature selection Support vector machine （SVM）
