基于K-Shape的时间序列模糊分类方法  

Fuzzy Classification for Time Series Data Based on K-Shape

作　　者：李海林[1,2] 贾瑞颖 谭观音[1,2] LI Hailin;JIA Ruiying;TAN Guanyin(Department of Information Management and Information Systems,Huaqiao University,Quanzhou Fujian 362021;Research Center for Applied Statistics and Big Data,Huaqiao University Xiamen Fujian 361021)

机构地区：[1]华侨大学信息管理与信息系统系,福建泉州362021 [2]华侨大学应用统计与大数据研究中心,福建厦门361021

出　　处：《电子科技大学学报》2021年第6期899-906,共8页Journal of University of Electronic Science and Technology of China

基　　金：国家自然科学基金面上项目(71771094);福建省自然科学基金面上项目(2019J01067);福建省社会科学规划一般项目(FJ2020B088)。

摘　　要：时间序列分类是数据挖掘中的重要主题,现有的大部分时间序列分类方法较少考虑到序列形状对分类结果的影响。该文提出了一种基于k-shape的时间序列模糊分类方法。该方法通过使用k-shape聚类算法对时间序列训练数据集各类别的成员进行聚类,获得各类别的聚类中心并形成聚类中心群,将每个类别的聚类中心群作为时间序列数据模糊分类的初始聚类中心,根据隶属度最大原则确定测试时间序列数据的类别标签。在30个时间序列公开数据集上的分类实验结果表明,该方法相较于SVM、Bayes、EAIW和TLCS这4种分类算法具有更好的分类性能,对具有扭曲和位移特征的时间序列数据分类有更好的可用性。Time series classification is an important topic in data mining.Most existing time series classification methods do not consider the influence of the shape of the time series on the classification results.The paper proposes a fuzzy classification method for time series based on k-shape.The method utilizes the k-shape clustering algorithm to cluster each category of the time series training datasets and obtains the cluster centers group of each class.After utilizing the cluster center group of each class as the initial clustering center of the fuzzy classification,class labels of the test datasets are determined according to the principle of maximum membership degree.Experimental results on 30 time series public datasets show that the proposed method has better classification performance than the traditional methods,including support vector machine(SVM),Bayes,ensemble algorithm of interval weightsc(EAIW),and trend information based on longest common subsequence(TLCS),with more excellent usability for time series with distortion and displacement characteristics.

关 键 词：分类算法 模糊分类 k-shape 时间序列 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]