不确定域环境下基于DKC值改进的K-means聚类算法

Improved K-means Clustering Algorithm Based on DKC in Uncertain Region Environment

提出一种不确定域环境下基于DKC值改进的K-means聚类算法,即U2d-Kmeans。该算法首先考虑到数据对象的不确定性因素,引入不确定域对数据对象进行描述;其次吸取2d-Kmeans的优点,对数据集进行预处理(剔除孤立点),并且采用累积距离的方法确定初始聚类中心,从而避免了随机选取聚类初始点造成聚类不稳定的缺陷;最后经过算法有效性对比实验证明得出,U2d-Kmeans算法比前两种算法更客观、有效。

This paper presented an improved K-means clustering algorithm based on DKC in uncertain region environment,namely U2d-Kmeans.Firstly,the algorithm takes uncertainty factors into account of the data object description,then uses new pretreatment method（removing isolated point） of data set and the cumulative distance method of determining the initial clustering center that is mentioned in the 2d-Kmeans algorithm.These methods avoid the defect of clustering instability caused by the random selection of clustering initial point.Finally,comparison experiment of the algorithm proves that the improved U2d-Kmeans is more objective and effective than the other two algorithms.