当前位置: X-MOL 学术Pers. Ubiquitous Comput. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
An improved fast shapelet selection algorithm and its application to pervasive EEG
Personal and Ubiquitous Computing ( IF 3.006 ) Pub Date : 2021-02-17 , DOI: 10.1007/s00779-020-01501-4
Xiunan Zou , Xiangwei Zheng , Cun Ji , Yuang Zhang

With the rapid development of pervasive devices, a great deal of time series are generated by various sensors, and many time series classification (TSC) algorithms have been proposed to deal with these data. Among them, shapelet-based algorithms have attracted great attention due to its high accuracy and strong interpretability. However, time complexity of shapelet-based algorithms is high. In this paper, we propose an improved Fast Shapelet Selection algorithm based on Clustering (FSSoC), which greatly reduces the time of shapelet selection. Firstly, time series are clustered into several groups with improved k-means, and then some time series are sampled from each cluster with a strategy based on Euclidean Distance sorting. Secondly, Important Data Points (IDPs) of the sampled time series are identified and only the subsequences between two nonadjacent IDPs are added to shapelet candidates. Therefore, the number of shapelet candidates is greatly reduced, which leads to a obviously reduction in time consumption. Thirdly, FSSoC is applied to shapelet transformation algorithm to test classification accuracy and running time, the experiments demonstrate that FSSoC is obviously faster than existing shapelet selection algorithms while keeping a high accuracy. At last, a case study on EEG time series is presented, which verifies the feasibility of FSSoC application to automatically discover representative EEG features.



中文翻译:

一种改进的快速小波选择算法及其在普适脑电中的应用

随着普及设备的快速发展,各种传感器产生了大量的时间序列,并且提出了许多时间序列分类(TSC)算法来处理这些数据。其中,基于shapelet的算法由于其准确性高和可解释性强而备受关注。但是,基于Shapelet的算法的时间复杂度很高。在本文中,我们提出了一种改进的基于聚类的快速小波选择算法(FSSoC),该算法大大减少了小波选择的时间。首先,将时间序列聚类为具有改进k均值的几组,然后使用基于欧氏距离排序的策略从每个聚类中采样一些时间序列。其次,确定采样时间序列的重要数据点(IDP),仅将两个不相邻IDP之间的子序列添加到小波候选中。因此,极大地减少了小形候选物的数量,从而明显减少了时间消耗。第三,将FSSoC应用于shapelet变换算法以测试分类精度和运行时间,实验表明,在保持高精度的同时,FSSoC明显比现有的shapelet选择算法快。最后,以脑电时间序列为例,验证了FSSoC应用自动发现具有代表性的脑电特征的可行性。第三,将FSSoC应用于shapelet变换算法以测试分类精度和运行时间,实验表明,在保持高精度的同时,FSSoC明显比现有的shapelet选择算法快。最后,以脑电时间序列为例,验证了FSSoC应用自动发现具有代表性的脑电特征的可行性。第三,将FSSoC应用于shapelet变换算法以测试分类精度和运行时间,实验表明,在保持高精度的同时,FSSoC明显比现有的shapelet选择算法快。最后,以脑电时间序列为例,验证了FSSoC应用自动发现具有代表性的脑电特征的可行性。

更新日期:2021-02-17
down
wechat
bug