一种基于密度峰值的半监督聚类算法

发布时间：2018-11-27 15:39

【摘要】：由于基于密度峰值的聚类算法对簇的形状不敏感,其聚类结果表现出良好的抗噪性。然而,当密度定义中变量难以反映簇的结构时,该算法性能下降明显,其主要原因在于聚类的非监督性。为此,该文在此算法的基础上提出了一种基于密度峰值的半监督聚类算法。该算法通过增加must-link和cannot-link约束作为先验知识,并在must-link约束集中叠加数据点的密度,以此产生新的聚类中心从而实现对数据点的吸引;对于cannot-link约束集中的数据点,通过将其n级最近邻居分离的方式找到其所属聚类中心,实现簇的归属。实验表明,基于密度峰值的半监督聚类算法利用先验知识来约束和引导聚类结果,在一定程度上改善了聚类的效果,并可应用于任意形状数据集的聚类问题中。
[Abstract]:Because the clustering algorithm based on the peak density is not sensitive to the shape of the cluster, the clustering results show good noise resistance. However, when the variables in the density definition can not reflect the structure of the cluster, the performance of the algorithm drops obviously, which is mainly due to the unsupervised nature of the clustering. Based on this algorithm, a semi-supervised clustering algorithm based on peak density is proposed in this paper. By adding must-link and cannot-link constraints as priori knowledge and superposing the density of data points in the must-link constraint set, the algorithm can generate new clustering centers to attract data points. For the data points in the cannot-link constraint set, the cluster center is found by separating the nearest neighbors of the class n, and the cluster attribution is realized. The experimental results show that the semi-supervised clustering algorithm based on peak density can use prior knowledge to constrain and guide the clustering results, improve the clustering effect to some extent, and can be applied to the clustering problem of arbitrary shape data sets.
【作者单位】：中南大学地球科学与信息物理学院;有色金属成矿预测与地质环境监测教育部重点实验室(中南大学);
【基金】：国家自然科学基金项目(41472301、41401532)
【分类号】：TP311.13

【相似文献】