一种基于离群点的聚类迭代检测算法

A Clustering Iterative Detection Algorithm Based on Outlier

大多数数据挖掘算法都可以对数据进行相对准确的分类,然而他们都集中于单独地使用聚类的方法。所以对于离群点存在的数据集,常常不能得出准确的结果。而COID算法(Cluster-outlier Iterative detection)把簇和离群点巧妙地结合起来,通过它们之间的关系来检测离群点并进行合理聚类。为进一步提高该算法的实用性,现利用prim算法确定初始簇中心,从而降低了迭代次数,实验证明改进后的算法具有更好的可行性、有效性和准确性,适合于高维数据中对于聚类检测的要求。

Most data mining algorithms can implement a relative accurate classification, but these algorithms only focus on clustering methods. Usually, these algorithms can＇ t get precise results when outliers exist in dataset. The COID algorithm （ Cluster-Outlier Iterative Detection） combines outliers with clusters, using the relationship be- tween them, to detect outliers and clustering reasonably. In order to further enhance the practicability of this algo- rithm, improvements on the basis of COID algorithm by using the prim algorithm is maken. This experiment shows that the improved algorithm can achieve better performance. It is valid for high dimension dataset and it can find outliers accurately and validly.