摘要
针对当前异常检测方法面临的分类性能有限以及分类结果易受噪声影响等问题,在分析当前异常检测方法的基础上,提出模糊大间隔最小超球模型FMHM。该模型引入模糊理论,在一定程度上减少了噪声对分类结果的影响;正常样本与奇异样本之间的间隔最大化确保错分率最小。标准UCI数据集上的比较实验表明,较之单类支持向量机(OCSVM)、支持向量数据描述(SVDD)、K近邻(KNN)等算法,所提方法 FMHM在异常检测方面具有一定优势。
In order to solve the problems of traditional outlier detection methods such as the limited efficiencies and noise-sensitiveness,this paper proposed fuzzy minimal hyper-sphere model with large margin（ FMHM） based on the analysis of traditional outlier detection methods. This proposed model introduced the fuzzy theory to decrease the influence of noises and tried to maximize the margin between the normal samples and the abnormal samples to ensure the misclassification efficiencies minimized. Comparative experiments with one class support vector machine（ OCSVM）,support vector data description（ SVDD）,K-nearest neighbor（ KNN） on the UCI datasets verify the effectiveness of FMHM in solving the problem of outlier detection.
出处
《计算机应用研究》
CSCD
北大核心
2017年第3期658-660,共3页
Application Research of Computers
基金
山西省自然科学基金资助项目(201601D011042)
山西省高等学校创新人才支持计划资助项目(2016)
关键词
异常检测
模糊理论
大间隔最小超球
outlier detection
fuzzy theory
minimal hyper-sphere model with large margin
