基于k近邻树的离群检测算法研究-计算机软件与理论专业论文.docxVIP

of the technology of detecting outliers are expounded. ③ A new dissimilarity measure method(dissimilarity measure method based on k nearest neighborhood) is proposed based on study of current dissimilarity measure method. The new method proposes that dissimilarity between data objects is directional for first time, and taking the density factors into account making the final algorithm more adapted to different density of data sets. ④ A new outlier detection method(Outlier Detection based on K-nearest Neighborhood MST，KNMOD) is present for the poor outlying clusters detection of current methods. The new algorithm greatly improves the detection efficiency of the algorithm with use of Fibonacci heap, and can overcome the problem of data sets shape with the help of tree features. Compared with other methods such as LOF,KNN,INFLO on the artificial data sets with obvious local outliers and outlying cluster proves the properties of KNMOD. At the same time, the comparison experiments with algorithm LOF, KNN, INFLO and COF on 5 real data sets from UCI is made and the validity and rationality of KNMOD are verified. Keywords: outlier detection, k nearest neighborhood, dissimilarity measure, minimum spanning tree, outlying cluster. 目 录 中文摘要 ............................................................................................................................... I 英文摘要 .............................................................................................................................. II HYPERLINK \l _bookmark0 1 绪 论............................................................................................................................. 1 HYPERLINK \l _bookmark1 1.1 研究背景及意义.............................................................................................................. 1 HYPERLINK \l _bookmark2 1.2 国内外研究现状.............................................................................................................. 2 HYPERLINK \l _bookma