基于空间平移的K-Means初始簇心选取  

K-Means Initial Cluster Center Selection Based on Spatial Translation

作　　者：朱家乐 

机构地区：[1]长安大学理学院,陕西 西安

出　　处：《应用数学进展》2024年第9期4381-4390,共10页Advances in Applied Mathematics

摘　　要：K-means聚类算法因其算法简单、计算效率高,在机器学习、数据挖掘等多个领域得到了广泛应用。然而,传统K-means算法在初始簇心的选取上存在随机性,这可能导致聚类结果的不稳定性。为了解决这一问题,本研究提出了一种基于空间平移的初始簇心选取算法。该算法首先将包含所有样本集的最小空间通过单位空间以一定步长遍历,在单位空间内统计样本点的密度,以此降低计算量。通过逐一选出密度最高的个点作为初始簇心,从而提高了K-means算法的聚类性能。在UCI的12种数据集上进行的实验表明,与传统的K-means、K-means++等算法相比,改进的算法在迭代次数上有所降低,聚类准确率得到了显著提高。K-means clustering algorithm is an important content in the field of machine learning and is widely used because of its simplicity and efficiency. In order to solve the problem that the initial cluster center selection of traditional K-means algorithm is random, an initial cluster center selection algorithm based on space segmentation is proposed. The minimum space containing all sample sets is divided to calculate the density, and the initial cluster centers with the highest density are selected one by one. The selected cluster centers are replaced by random initial cluster centers for K-means clustering. Twelve datasets were tested separately at UCI. The experimental results show that compared with traditional K-means, K-means++ and other algorithms, the improved algorithm has lower iteration times and higher clustering accuracy.

关 键 词：K-MEANS 初始聚类中心 密度 空间平移 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]