基于加权FCM聚类算法的特征提取与降维研究  

Feature Extraction and Dimension Reduction Based on Weighted FCM Clustering Algorithm

作　　者：邱林润 李蓉蓉[1] QIU Linrun;LI Rongrong(Guangdong University of Science and Technology,Dongguan 523000,Chin)

机构地区：[1]广东科技学院,广东东莞523000

出　　处：《电视技术》2018年第6期24-27,共4页Video Engineering

基　　金：省级课题:2015年广东省高校重大科研项目(青年创新人才项目):基于移动互联网的一体化掌上校园的研究与实现项目编号:2015KQNCX188

摘　　要：本文所涉及的降维思维是基于聚类过程和数据相似性产生的,界标等距映射算法在FCM-LI算法当中起到了至关重要的作用,在FCM中间结果对特征向量进行重新提取,尽可能减少特征向量维数,简单化处理FCM计算过程。结果表明,FCM-LI算法与传统高维数据直接分类算法相比,前者的优势比较体现在能够快速完成聚类过程。FCM-LI和FCM的差别在维数降到11维时约为3.95%,然而此时需消耗更多的时间来运行。将维数控制在5维时,此时运行时间最短,但不能确保其准确度,也意味着如果以过低的维数运行则将导致原数据出现错误,无法得到准确的分类结果。In this paper, a kind of feature extraction is carried out by weighted FCM clustering algorithm in the process of clustering, and the idea of using the similarity reduction of data is used. The FCM - LI algorithm uses the boundary mapping algorithm to extract feature vectors from the intermediate result of FCM. The aim is to reduce the dimension of the feature vector, speed up the speed of clustering and reduce the speed of clustering. The computational complexity of the traditional FCM. The results show that FCM -LI has faster clustering speed than traditional clustering algorithm based on high - dimensional data. When the dinlension reduction is small, the difference between FCM and FCM -LI is very small. When the dimensinn is reduced to 11 dimension, the difference between FCM - LI and FCM is about 3.95% , but the running time is greatly improved. When the dimension is reduced to 5 dimension, the running time is very short, but the accuracy is not guaranteed. It shows that the dimension is too low, which leads to the distortion of the original data structure, and it is very rare to be classified in a corrcet way.

关 键 词：FCM算法 聚类 FCM-LI算法 特征 降维 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]