CAJ | 학술논문

为解决K‐M eans算法对初始聚类中心的敏感性，提出基于精细簇的K‐M eans稳定算法。采用基于密度算法的预分类技术来获得高密度的核心类，确定能够代表数据集结构的类层次树；根据类层次树中具有较高代表性的子类中心进行K‐M eans聚类，获得精细簇；按照层次树中的类归属合并精细簇，获得精确稳定的聚类效果。实验结果表明，该方法能够克服K‐M eans由于随机初始质心造成的结果不稳定性，一定程度上提高了聚类效果。
위해결K‐M eans산법대초시취류중심적민감성，제출기우정세족적K‐M eans은정산법。채용기우밀도산법적예분류기술래획득고밀도적핵심류，학정능구대표수거집결구적류층차수；근거류층차수중구유교고대표성적자류중심진행K‐M eans취류，획득정세족；안조층차수중적류귀속합병정세족，획득정학은정적취류효과。실험결과표명，해방법능구극복K‐M eans유우수궤초시질심조성적결과불은정성，일정정도상제고료취류효과。
To decrease the sensitivity of initial centroids of K‐Means ,a stable method based on refined cluster was proposed .The density based algorithm was used for pre‐classification to get CHT (cluster‐hierarchy‐tree) that contained the cluster structure of data set .The sub‐clusters with high density in CHT were chosen as initial parameters of K‐Means .According to the relationship in CHT ,the small classes produced by K‐Means were merged and the clustering result was gained .The experimental results show the stability and better clustering performance of the proposed method .