SE-shapelets: Semi-supervised Clustering of Time Series Using Representative Shapelets
This addresses the challenge of improving clustering accuracy for time series data, particularly in semi-supervised settings, though it appears incremental as it builds on existing shapelet-based methods.
The paper tackles the problem of low clustering accuracy in time series due to uninformative subsequences by proposing SE-Shapelets, a semi-supervised method that uses labeled and pseudo-labeled data to discover representative shapelets, achieving higher clustering accuracy than counterpart methods on UCR datasets.
Shapelets that discriminate time series using local features (subsequences) are promising for time series clustering. Existing time series clustering methods may fail to capture representative shapelets because they discover shapelets from a large pool of uninformative subsequences, and thus result in low clustering accuracy. This paper proposes a Semi-supervised Clustering of Time Series Using Representative Shapelets (SE-Shapelets) method, which utilizes a small number of labeled and propagated pseudo-labeled time series to help discover representative shapelets, thereby improving the clustering accuracy. In SE-Shapelets, we propose two techniques to discover representative shapelets for the effective clustering of time series. 1) A \textit{salient subsequence chain} ($SSC$) that can extract salient subsequences (as candidate shapelets) of a labeled/pseudo-labeled time series, which helps remove massive uninformative subsequences from the pool. 2) A \textit{linear discriminant selection} ($LDS$) algorithm to identify shapelets that can capture representative local features of time series in different classes, for convenient clustering. Experiments on UCR time series datasets demonstrate that SE-shapelets discovers representative shapelets and achieves higher clustering accuracy than counterpart semi-supervised time series clustering methods.