关键词: 聚类，模糊均值，属性加权，密度，误分类数

Abstract: The traditional clustering algorithm will converge to a local minimum point when the initial objects' attributes have no obvious difference,which can cause the decline of algorithms' accuracy and incorrectness of the results. In order to overcome these drawbacks, a density based weighted fuzzyc一 mean clustering algorithm was proposed. It used the results of the calculation of the relative density differences attributes to determine the initial partition. After obtaining the better initial centers,a weighted fuzzy algorithm which can distinguish the importance of each attribute was implemented. Experimental results show that the algorithm not only can discriminate the attributes’contribution, but also can improve the stability and accuracy.

Key words: Clustering, Fuzzy C-means, Attribute weighted, Density, Number of misclassification