Efficient class-specific shapelets learning for interpretable time series classification

Time series classification is a critical problem in data mining, with many applications in diverse areas, including medical, biological, financial, engineering, and industrial [1]. As a result, hundreds of algorithms [2] are proposed to tackle this problem. Among those methods, shapelet-based approaches [3-5] have attracted increasing attention in recent years due to the comprehensive interpretability of the classification results. Shapelet-based methods discover discriminatory subsequences from the raw time series and take the similarity between the time series and the shapelets as features to build the classifier. The shapelets represent distinguishing local patterns of some classes [3] thus a classification decision can be directly made by the presence or absence of particular shapelets. Despite the significant advantages, most existing shapelet-based methods discover shapelets by checking all possible In the last decade, time series classification approaches based on time-independent shape lets, have received considerable attention due to their high prediction accuracy and intuitive interpretability. However, most existing shapelets discovery approaches find shapelets by evaluating the discriminatory power of all subsequences of the series, which is computationally expensive even with certain speed-up techniques. Even though some shapelet learning approaches learn the near-to-optimal shapelets from the training series rather than searching from numerous segments, they still have significant drawbacks in their performance regarding the accuracy, efficiency, and interpretability due to the numerous class-shared shapelets with fixed lengths. Thus, we propose a new shapelet learning approach that can learn as few as possible class-specific and variable-length shapelets. Extensive experiments demonstrate that our proposed method is competitive about classification accuracy over 18 baselines on 25 datasets, outperforms 2 orders of magnitude about efficiency, and is more interpretable than existing classifiers. (c) 2021 Elsevier Inc. All rights reserved.

Automated summary

Time series classification is a critical problem in data mining with applications in various fields, leading to the proposal of numerous algorithms. Shapelet-based approaches have gained attention for their interpretability in classification results. These methods discover discriminatory subsequences and use similarity between time series and shapelets as features for building classifiers, enabling direct classification decisions based on the presence or absence of specific shapelets.