Abstract: The basic idea of Self-Organizing Feature Maps（SOFM） algorithm and Clustering LARge Applications（CLARA） algorithm is introduced.Firstly this paper adopts the SOFM algorithm clustering on the data set to attain the class number and the neurons’s connection weights vector，and then the clustering result is used to initialize the class number and the k medoid in the CLARA algorithm.The simulate experiment shows that this clustering method can improve the clustering performance.

Key words: Self-Organizing Feature Maps（SOFM）, Clustering LARge Applications（CLARA）, clustering, replace cost

摘要： 介绍了自组织特征映射（SOFM）算法及大规模应用聚类（CLARA）算法的基本思想，提出了一种首先利用SOFM算法对数据集进行粗聚类，确定簇的数目k和神经元的连接权向量，然后从数据集中找出与SOFM算法的神经元的连接权向量最相似的k个代表点作为CLARA算法的k个代表点的初始值的改进CLARA算法。实验结果表明，改进算法具有更高的聚类效率和更好的聚类质量。

关键词: 自组织特征映射, 大规模应用聚类, 聚类, 替换代价