自然最近邻密度聚类算法的改进方法  

An Improved Method to Clustering Algorithm of Natural Nearest Neighbor Density

作　　者：李俊海[1] LI Junhai(College of Science,Henan University of Technology,Zhengzhou 450007,China)

机构地区：[1]河南工业大学理学院,河南郑州450001

出　　处：《新乡学院学报》2020年第12期38-42,共5页Journal of Xinxiang University

基　　金：河南省高等学校重点科研项目(20B416001)。

摘　　要：自然最近邻算法(TNDP)所得结果的类内差异很小,若待分类集合元素较多,则可能出现过度分割问题——具有相似特征元素的子集被分割为多个小类。为此,在使用TNDP算法后,若类内平方和差分小于给定的阀值,就把距离最近的两个子类合并为一个大类,如此反复,直到类间的距离足够大为止。这可保证“具有相似特征元素的多个子类”合并为一个较大类,从而提高最终分类结果的可解释性。Since the results of the natural nearest neighbor algorithm(TNDP)have small intra-class differences,if there are many elements in the set to be classified,there may be an over-segmentation problem,a subset of elements with similar characteristics is divided into multiple small classes.This article suggests that after using the TNDP algorithm,if the difference of the sum of squares within the classes is less than a given threshold,the closest two subclasses can be combined into a new big class.Do this until the distance between each class is large enough.This can often ensure that"multiple classes with similar feature elements"are merged into one larger category,thereby improving the interpretability of the final classification result.

关 键 词：自然最近邻居 密度聚类算法 类间相似度 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]