摘要
k最近邻孤立点检测算法的检测结果受用户设置参数的影响较大,并且无法判定孤立点强弱,针对该缺陷,引入阈半径和密集度阈值,提出基于最近邻距离差的孤立点检测算法。通过在多个数据集上的实验表明,改进算法扩大了参数的设置范围,降低了参数对结果的影响,并能够有效检测出强孤立点,用户通过调整密集度阈值,可以判定孤立点强弱,改进算法增强了原算法的稳定性和灵活性。
Results of k nearest neighbor outlier detection algorithm are affected by parameters set by users deeply and are unable to determine the strength. In order to eliminate this defect, the threshold radius and density threshold is introduced and improved outlier detection algorithm is presented based on difference between nearest neighbors distance. The experimental results of sev- eral data sets show that improved algorithm extends the span of parameters, reduces the impact of parameters on the results and can effectively detect strong outliers. By setting the intensity threshold, users can determine the strength of the outliers. The improved algorithm enhances the stability and flexibility of the original algorithm.
出处
《计算机工程与设计》
CSCD
北大核心
2013年第4期1265-1269,共5页
Computer Engineering and Design
基金
国家自然科学基金项目(60970059
61170136)
山西省自然基金项目(2011011015-4)
山西省青年基金项目(2011021013-3)
太原理工大学校青年基金项目(K201021)
关键词
孤立点检测
最近邻距离差
参数设置
k最近邻
强孤立点
outlier detection
difference between nearest neighbors distance~ parameter settings~ k nearest neighbor~ strong outliers