Large-Scale Time Series Clustering Based on Fuzzy Granulation and Collaboration

Abstract

Clustering a group of large-scale time series with the same length is a frequently met problem in real world. However, the existing clustering methods often show high computational cost and low efficiency when dealing with this problem. In this paper, we propose a granulation-based horizontal collaborative fuzzy clustering method for this problem. In this method, some new subgroups are built from the given large-scale time series group by segmenting all the time series with time alignment. Thus, all the subsequences in each subgroup have the same length, which are much smaller than the length of the original time series. Just because of the smaller length, the clustering of all subsequences in each subgroup can be easily carried out with lower computation cost and higher efficiency. How to aggregate the clustering results of all subgroups to obtain the clustering result of the given group of large-scale time series becomes the main task of this paper. To solve this problem, we carry out the horizontal collaborative fuzzy clustering on the last subgroup by collaborating the clustering information of the previous subgroups. To obtain much better performance, we first perform fuzzy information granulation on the original group of large-scale time series. After that, the original group of time series is transformed into a group of granular time series. Therefore, the collaborative clustering is carried out on the corresponding granular subgroups. Simulation experiments presented here illustrate the good performance of the algorithm of our method. C 2015 Wiley Periodicals, Inc.

Autoregressive mixture models for clustering time series

“…Alternatives to hierarchical cluster analysis include using community network detection and quasi U‐statistics (Valk and Pinheiro, 2012; Ferreira and Zhao, 2016). Soft clustering includes centroid based techniques and mixture models and can identify useful mixed memberships, which is often preferred over hard clustering in situations with heterogeneous data (D'Urso and Maharaj, 2009; Genolini et al, 2015; Wang et al, 2015).…”

Kernel‐based low‐rank tensorized multiview spectral clustering

“…Besides, clustering has always been an essential task in data mining. It has a wide range of applications in many fields, such as medical diagnosis, 2 image processing, 3,4 grouping in crowd evaluation, 5,6 social network analysis, 7 time series analysis, 8 and so on. Aiming to partition data into different clusters with multiple views, multiview clustering is an important branch of clustering.…”