Scalable Clustering of Time Series with U-Shapelets

Abstract

A recently introduced primitive for time series data mining, unsupervised shapelets (u-shapelets), has demonstrated significant potential for time series clustering. In contrast to approaches that consider the entire time series to compute pairwise similarities, the u-shapelets technique allows considering only relevant subsequences of time series. Moreover, u-shapelets allow us to bypass the apparent chicken-and-egg paradox of defining relevant with reference to the clustering itself. U-shapelets have several advantages over rival methods. First, they are defined even when the time series are of different lengths; for example, they allow clustering datasets containing a mixture of single heartbeats and multi-beat ECG recordings. Second, u-shapelets mitigate sensitivity to irrelevant data such as noise, spikes, dropouts, etc. Finally, u-shapelets demonstrated ability to provide additional insights into the data. Unfortunately, the state-ofthe-art algorithms for u-shapelets search are intractable and so their advantages have only been demonstrated on tiny datasets. We propose a simple approach to speed up a ushapelet discovery by two orders of magnitude, without any significant loss in clustering quality.

Discovering patterns of online popularity from time series

“…There have been partitional [30,43,33,12], subsequence matching [6,22], hierarchical [21] approaches using these different time series distance measures. Moreover, a major body of work [38,47] exists in subsequence matching based time series clustering where they identify shorter most identifying portions of time series data also known as shapelets to group them. For the multivariate time series data, same categorizations can be made as modeling based [17,14], and variants of generalizing univariate solutions to multivariate cases [36,13].…”

“…A wide range of studies have been focused on solving the problem of efficiently clustering time series [2]. Some approaches have adopted methods such as subsequence techniques [38,46,47], and dimensionality reduction models [45,7]. However, we aim at uncovering different patterns of popularity on the basis of both their shapes and the time at which the main popularity gain is achieved.…”

Speeding Up All-Pairwise Dynamic Time Warping Matrix Calculation

“…Recently, the community has devoted some effort to improve the efficiency of time series clustering algorithms. For instance, Zhu et al have framed the problem of calculating the all-pairwise DTW matrix as an anytime algorithm and applied such approach to clustering [13] and Ulanova et al have proposed new techniques to cluster temporal objects in admissible time [11].…”

Enabling process mining on sensor data from smart products

“…1, which would first need to be segmented and labelled. Another approach that has been shown to be able to discover activities in segmented sensor data is that of clustering using U-Shapelets [16]. This approach clusters segments of sensor data based on the characterising shape of the data within a segment.…”

“…The number of clusters was determined iteratively and based on interpretation of the cluster centroids using domain knowledge. The use of U-Shapelet clustering was also investigated, as it has been used in other studies with similar data to discover different types of activities from sensor data [16]. However, evaluation on the labelled dataset showed that this technique does not separate different activities into different clusters.…”