摘要
对不平衡数据进行聚类分析时,K-means聚类方法可能会错误地将分布在较小区域类别中的样本划分到大区域类别中;谱聚类算法,虽然可以有效优化数据结构,并很好地识别不同形状的样本,但却难以处理大规模数据.针对这些问题,提出一种改进地标点采样的不平衡数据聚类算法.该算法首先对不平衡数据进行预聚类以获得初始类标签,然后基于数据密度对数据进行采样.在此基础上,通过对采样数据执行K-means聚类,并将聚类中心作为地标点,对数据进行谱聚类分析.实验结果显示,该方法在处理不平衡数据时,不仅能够有效提高样本的聚类准确率,而且能够保证聚类结果的稳定性和精度.
When clustering imbalanced data,K-means clustering method can easily misclassify samples distributed in smaller regions into larger regions.While spectral clustering algorithm can effectively optimize the data structure and better identify samples of different shapes,it is difficult to deal with large-scale data.To solve these problems,this paper introduces an improved imbalanced data clustering algorithm with improved landmark sampling.The algorithm first pre-clusters the imbalanced data to obtain the initial class label,then samples the data based on its density distribution and applies K-means clustering on the sample.Finally,it selects the cluster center as the landmark for spectral clustering analysis.The experimental results show that when dealing with imbalanced data,this method can not only improve the clustering accuracy of the sample,but also ensure the stability and accuracy of clustering results.
作者
韩素青
李淑慧
HAN Suqing;LI Shuhui(Department of Computer Science and Technology,Taiyuan Normal University,Jinzhong 030619,China)
出处
《太原师范学院学报(自然科学版)》
2019年第2期34-39,共6页
Journal of Taiyuan Normal University:Natural Science Edition
基金
山西省重点研发计划项目:基于GIS的智能消防远程预警平台研究(201803D121088)
关键词
不平衡数据
谱聚类
地标点采样
奇异值分解
imbalanced data
spectral clustering
landmark sampling
singular value decomposition