Clustering Time Series Using Unsupervised-Shapelets

Abstract

Abstract-Time series clustering has become an increasingly important research topic over the past decade. Most existing methods for time series clustering rely on distances calculated from the entire raw data using the Euclidean distance or Dynamic Time Warping distance as the distance measure. However, the presence of significant noise, dropouts, or extraneous data can greatly limit the accuracy of clustering in this domain. Moreover, for most real world problems, we cannot expect objects from the same class to be equal in length. As a consequence, most work on time series clustering only considers the clustering of individual time series "behaviors," e.g., individual heart beats or individual gait cycles, and contrives the time series in some way to make them all equal in length. However, contriving the data in such a way is often a harder problem than the clustering itself.In this work, we show that by using only some local patterns and deliberately ignoring the rest of the data, we can mitigate the above problems and cluster time series of different lengths, i.e., cluster one heartbeat with multiple heartbeats. To achieve this we exploit and extend a recently introduced concept in time series data mining called shapelets. Unlike existing work, our work demonstrates for the first time the unintuitive fact that shapelets can be learned from unlabeled time series. We show, with extensive empirical evaluation in diverse domains, that our method is more accurate than existing methods. Moreover, in addition to accurate clustering results, we show that our work also has the potential to give insights into the domains to which it is applied.

Scalable Clustering of Time Series with U-Shapelets

“…Although u-shapelets have shown considerable promise for time series clustering, the u-shapelet extraction algorithms proposed to date are intractable for large datasets [26]. To mitigate this, the algorithm in [26] resorts to computing gap scores for the subsequences of just the first time series in the dataset, making it order dependent and brittle to an unusual instance being the first item encountered.…”

“…To mitigate this, the algorithm in [26] resorts to computing gap scores for the subsequences of just the first time series in the dataset, making it order dependent and brittle to an unusual instance being the first item encountered. To eliminate this undesirable property, we must compute the gap score for every subsequence of each time series in the dataset.…”

“…Virtually all research has focused on introducing novel similarity measures and/or novel clustering techniques. In contrast, a recent technique, u-shapelets [26], uses the Euclidean distance as the similarity measure, and a k-meanslike technique as the clustering algorithm. The novelty of ushapelets is in selectively ignoring most of the data, and only using a small number of subsequences for clustering.…”

fMRI Data Analysis Using Dempster-Shafer Method with Estimating Voxel Selectivity by Belief Measure

“…Recently, clustering time series [14,15] has newly concerned a considrable amount of research. They have shown their efficiency because they form partitions based on similarity of voxel values in the fMRI time series where each partition is represented by the cluster centroid that is sufficient for the analysis and investigation of fMRI time series [16].…”

Battery Grouping with Time Series Clustering Based on Affinity Propagation

“…Another similar problem for time series clustering is time series classification, which classifies time series to different classes and is a supervised learning problem, while the time series clustering problem is an unsupervised problem. Commonly used methods for time series classification are the shapelet based method [15], the recurrent neural network (RNN) based method [16], etc. In fact, sometimes we can use the time series classification method to accomplish time series clustering.…”