子空间聚类算法在高维数据异常检测中的应用  

Application of subspace clustering algorithm in anomaly detection of high-dimensional data

作　　者：周冬[1] 苏勇[1] 黄烨[1] 

机构地区：[1]江苏科技大学计算机科学与工程学院,江苏镇江212003

出　　处：《信息技术》2013年第3期168-171,共4页Information Technology

摘　　要：传统异常检测技术是基于距离和密度的,快速的异常检测算法过分依赖于索引结构或网格划分,在低维数据上有很好的效果;面对高维数据的稀疏性、空空间现象等特性,索引结构失效,网格划分的数目呈指数级增长,传统算法性能下降;文中采用信息熵确定高维数据异常子空间,在异常子空间上使用DBSCAN聚类算法,在高维数据异常检测中表现出较好的性能。Traditional anomaly detection is based on the distance and density, fast anomaly detection algorithm is too dependent on the index structure and meshing, having a good effect on the low-dimensional data, due to the sparsity of high-dimensional data, the empty space phenomenon and so on, the index of structural lose its meaning, the number of meshing get a exponential growth, the performance of traditional algorithms declined, the article presents a algorithm which determining the anomaly subspace of the high-dimensional data, using DBSCAN clustering algorithm in the anomaly subspace, show a better performance in the anomaly detection of high-dimensional data.

关 键 词：高维数据 信息熵 异常检测 子空间聚类算法 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]