摘要
针对现有的无监督异常检测技术的不足之处,提出了一种基于样本分布异常数据实例度量方法。针对数据对象是高维数据的问题,将主成份分析方法应用到异常检测中解决数据集的降维问题。在此基础上,提出了一种新的无监督异常检测算法μ UAD,并对该算法进行了性能评估。
Aiming at the weaknesses of current unsupervised anomaly detection techniques,a measurement approach about samples distribution of anomaly data is proposed.Aiming at the problem that dataset is of high dimensions,principal components analysis is used in anomaly detection to reduce the dimensions of dataset.Therefore a novel unsupervised anomaly detection approach μ-UAD is described.The performance of the approach is evaluated.
出处
《信息工程大学学报》
2004年第3期56-59,共4页
Journal of Information Engineering University
关键词
异常检测
无监督学习
聚类
主成份分析
anomaly detection
unsupervised learning
clustering
principal components analysis