基于聚类融合的混合属性数据增量聚类算法  

Incremental clustering algorithm of mixed numerical and categorical data based on clustering ensemble

作　　者：李桃迎[1] 陈燕[1] 张金松[1] 秦胜君[1] 

机构地区：[1]大连海事大学交通运输管理学院,辽宁大连116026

出　　处：《控制与决策》2012年第4期603-608,共6页Control and Decision

基　　金：国家自然科学基金项目(70940008);高等学校博士学科点专项科研基金项目(200801510001)

摘　　要：针对传统增量聚类方法对混合属性数据聚类时存在不稳定、随机性大和准确性不够高的缺点,提出一种基于聚类融合的混合属性数据增量聚类算法.该算法以传统增量聚类为基础,采用多种聚类算法的结果进行融合来代替原有单一划分,并重新修正了阈值的取值范围.实验表明,所提出的算法利用原有数据的特征,提高了聚类的稳定性和精确性,具有很好的聚类效果.Traditional clustering methods have disadvantages of unsteadiness, randomness and low accuracy for classifying mixed numerical and categorical data. Therefore, the incremental clustering algorithm of mixed numerical and categorical data based on clustering ensemble is proposed, which adopts the results of several clustering to replace that of single clustering and modifies the design of threshold. The experiment results show that the improved algorithm has higher stability and accuracy by using the characters of existing data, and possess better effectiveness.

关 键 词：聚类融合 增量聚类 矢量量化 模糊κ-均值 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]