不均衡数据集下基于SVM的托攻击检测方法

Shilling Attack Detection Method Based on SVM Under Unbalanced Datasets

传统支持向量机(SVM)方法在数据不均衡情况下无法有效实现托攻击检测。在研究SVM的基础上,提出一种基于欠采样和代价敏感SVM相结合的托攻击检测方法。利用边界样本修剪技术实现训练样本的均衡,在消除部分多数类样本显著减小数据不均衡程度的同时,保证信息损失最小。结合受试者工作特征分析技术,利用代价敏感SVM对重构后的样本集进行训练,在限定范围内自动搜索最优参数,进而调节阈值获得系统决策函数。实验结果表明,该方法能提高托攻击的检测精度。

Traditional Support Vector Machine（SVM） drops significantly when it is applied to the problem of learning from unbalanced datasets. Based on the study of SVA, a new classifying method which combines the method of under-sampling and cost-sensitive SVM together is proposed. In the first stage, balanced data are set by reconstructing both the majority and the minority class. And in the second stage, cost sensitive SVM is conducted for detection decision function. Receiver Operating Characteristic（ROC） analysis is used to select optimum parameters of cost sensitive SVM in limited grid scope. The proposed model is used for attack detection on recommender systems. Experimental results show that the proposed method can improve the classification accuracy.