一种基于属性相似度的孤立点挖掘方法

AN OUTLIER MINING ALGORITHM BASED ON ATTRIBUTE SIMILARITY

孤立点挖掘是数据挖掘中研究的热点之一。在对已有的孤立点挖掘技术分析的基础上,结合基于密度的聚类算法,提出了一种新的改进的检测孤立点方法即基于属性相似度的孤立点挖掘方法(ADBSCAN)。该方法先用基于密度的聚类算法进行聚类,然后再利用对象间的属性相似度进行进一步的检验,确定不包含在任何聚类中的对象是否为真正的孤立点,并通过实验验证了该方法的可行性和有效性。

Outlier mining is one of the research focuses in data mining. Based on the analysis of existing outlier mining technology,and in conjunction with the density-based clustering algorithm,we put forward a new improved outlier detection algorithm which is called outlier mining based on attribute similarity （ ADBSCAN） . It clusters with density-based clustering algorithm firstly,and then makes further detection using the similarities between objects to determine whether or not an object out of any cluster is a real outlier. The feasibility and the effectiveness of the new algorithm have been attested by the experiment.