基于竞争学习的K质心组合聚类算法

Combination Clustering Algorithm of K-Centroid Based on Competitive Learning

K-Means算法的聚类结果对初始簇的选择非常敏感,通常获得的是局部最优解而非全局最优解。为此,在K-Means聚类算法基础上,引入组合聚类和竞争学习概念,提出一种基于竞争学习的K质心组合聚类算法CLK-Centroid。该算法采用竞争学习策略计算簇的质心,以适应噪声数据和分布异常数据的要求,使用组合聚类策略提高聚类的精度。在数据集上构建多个CLK-Centroid聚类器进行聚类,构建子簇相似矩阵,并根据子簇之间的相似性合并相似簇。理论分析和实验结果表明该算法能够提高聚类质量。

For the choice of initial cluster K-Means clustering algorithm is very sensitive, its results are frequently a local optimal solution rather than a global optimal solution. On the premise of studying K-Means clustering algorithm, this paper introduces a concept of clustering and competitive learning, and proposes a combination clustering algorithm of K-Centroid based on competitive learning（CLK-Centroid algorithm）. CLK-Centroid algorithm adapts the noise data and outliers by using a strategy of competitive learning to compute a cluster Centroid, and improves the precision of cluster by using a strategy of combination clustering. Building a multiple of CLK-Centroid clustering models to cluster on data set, the different sub-cluster that comes from the different clustering results must contain an intersection. The sub-cluster similarity matrix is built to merge similar cluster according to the similarity between the sub-clusters. Theoretical analysis and experimental results show that the algorithm can improve the clustering quality.