基于层次策略的半监督K-medoids算法研究

针对现有基于K-means的半监督聚类算法存在的共同问题,即对离群点敏感、在非凸数据集与不平衡数据集上表现差,提出了一种基于层次策略的散布种子半监督中心聚类算法。首先通过基于影响空间的样本边缘因子将数据集分为核心层与边缘层,然后应用一种改进的K-medoids算法完成核心层聚类,最后采用一种递进半监督分配策略对边缘层进行分配得到最终聚类结果。算法通过层次策略解决了离群点干扰问题、半监督子簇聚类及合并策略实现了在不同分布数据集上有效聚类。通过与几种半监督聚类方法在人工数据集以及真实数据集上进行的对比实验证明,该算法能够解决现存问题,提升了聚类性能与鲁棒性。

Research on semi-supervised K-medoids algorithm based on hierarchical strategy

Aiming at the common problems of existing semi-supervised clustering algorithms based on K-means,such as sensitivity to outliers,poor performance in non-convex data sets and unbalanced data sets,this paper proposed a hierarchical strategy based semi-supervised clustering algorithm for seed dispersal.Firstly,it divided the dataset into the core layer and the border layer by strategy based on the border factor of the influence space.Then,it used the improved K-medoids algorithm to complete the core layer clustering.Finally,it used a progressive semi-supervised distribution strategy to allocate the border layer and obtained the final clustering results.The algorithm solved the outlier interference problem,semi-supervised sub-cluster clustering and merge strategy,and achieved effective clustering on different distributed datasets.Comparing experimental results on artificial datasets and real datasets show that the algorithm can solve the existing problems and has better clustering performance and robustness.