基于shapelets学习的多元时间序列分类

多元时间序列广泛存在于日常生活中的各个领域,多元时间序列分类是从时间序列数据中获取信息的基本方法。目前,时间序列分类研究面临着相似性度量方法特殊、原始数据维度高等问题,现有的多元时间序列分类方法的分类性能仍有待提高。文中提出一种基于shapelets学习的多元时间序列分类方法。首先,提出了新的正则化最小二乘损失学习框架下的shapelets学习方法,在此基础上采用基于shapelets的一元时间序列分类方法对多元时间序列的每维一元数据进行分类,随后由各维上的分类结果投票决定多元时间序列的最终分类结果。实验证明,所提方法在多元时间序列分类问题中能够取得较高的分类精度。

Multivariate Time Series Classification Based on Shapelets Learning

Multivariate time series data exist in a wide range of real-life domains,and multivariate time series classification is a basic method of obtaining information from time series data.At present,time series classification is suffered from the problem that the similarity measure of time series data is special and the dimension of the original data is high,thus the classification performance of the existing multivariate time series classification methods still need to be improved.This paper presented a multivariate time series classification method based on shapelet learning.At first,this paper established a shapelets learning method under a regularized least squares loss learning framework,and the time series classification method with one dimension based on shapelets is used to classify the vrivariate data of multivariate time series.Then the final resut of the multivariate time series is determined through plurality voting.Experimental results indicate that the proposed method achieves high classification accuracy when processing multivariate time series classification problem.