函数型数据聚类分析研究综述与展望  

Review and Prospect of Functional Data Clustering Analysis

作　　者：王德青[1,2] 朱建平[2] 刘晓葳[2,3] 何凌云[1] 

机构地区：[1]中国矿业大学管理科学与工程博士后流动站,江苏徐州221116 [2]厦门大学数据挖掘研究中心,福建厦门361005 [3]华侨大学数量经济研究院,福建厦门361021

出　　处：《数理统计与管理》2018年第1期51-63,共13页Journal of Applied Statistics and Management

基　　金：教育部人文社会科学研究青年基金(15YJCZH162);全国统计科学研究项目(2016LY13,2016LZ13);中国博士后科学基金面上资助(2015M571839);中央高校基本科研业务费专项基金(2015WA01)的阶段性研究成果;湖南省社会科学基金(15YBA085)资助

摘　　要：函数型数据是大数据时代的典型数据,也是大数据分析的重要视角,其稀疏粗糙、无穷维、低信噪比等复杂特性导致传统聚类分析方法凸显诸多弊端。为了厘清函数型数据聚类分析的研究现状,在界定函数型数据概念与内涵基础上,本文依据方法原理差异将函数型数据聚类分析方法划分为四类,理论剖析并模拟检验每一类别方法的相对优势和存在的不足。最后,针对现有研究尚待解决的关键问题,并结合大数据时代的数据特征,展望了函数型数据聚类分析的未来研究方向。Functional data is not only a typical data in big data era, but also a prospective of big data analysis, its sparsity, roughness, infinite dimension, low signal-noise ratio and other complex character- istics lead to many limitations of traditional clustering analysis methods derived from discrete data. In order to clarify the research status of functional data clustering analysis, under definition of functional data and its connotation, this paper firstly divides function data clustering methods into four groups based on different principles of methods. Secondly, this paper presents a comprehensive comparison on relative advantage and deficiency of each group through theoretical analysis and simulation tests. At last, combining the key problems unsolved in the present study with the data characteristics of the big data era, the future research direction of functional data clustering analysis is discussed.

关 键 词：函数型数据 聚类分析 文献综述 研究展望 

分 类 号：C812[社会学—统计学] O212[理学—概率论与数理统计]