CAJ | 학술논문

传统的K-means算法只能识别出类似球形的数据集，传统的MST聚类算法虽然能识别任意形状的数据集，但对噪声和异常点十分敏感，由此提出了一种将K-means聚类与MST聚类相结合的聚类算法。此算法先使用K-means算法将数据分割成多个小的类似球形的数据集，然后对各个小的数据集的均值点采用MST聚类算法进行聚类分析。实验证明此算法具有较好的抗干扰性，并且可以识别出任意形状分布的数据集。为了评价聚类算法的性能，文中同时提出了一种新的聚类质量评价函数，实验证明此评价函数是有效的。
전통적K-means산법지능식별출유사구형적수거집，전통적MST취류산법수연능식별임의형상적수거집，단대조성화이상점십분민감，유차제출료일충장K-means취류여MST취류상결합적취류산법。차산법선사용K-means산법장수거분할성다개소적유사구형적수거집，연후대각개소적수거집적균치점채용MST취류산법진행취류분석。실험증명차산법구유교호적항간우성，병차가이식별출임의형상분포적수거집。위료평개취류산법적성능，문중동시제출료일충신적취류질량평개함수，실험증명차평개함수시유효적。
Classical K-means only can distinguish similar sphere-cluster, classical MST clustering algorithm can distinguish arbitrary cluster, but it's very sensitive to noisy and abnormal data, the paper combine K-means and MST clustering algorithm. Firstly, K-means algorithm divides data to many similar mini sphere-clusters, and then MST clustering algorithm deals with mean points of these mini sphere-clusters. Results show that the new algorithm has good anti-noisy ability, and can distinguish arbitrary cluster. In order to an-alyze the quality of clustering, the paper provides a new quality evaluation function. Results show the func-tion is effective.