On the Convergence of Symbolic Pattern Forests and Silhouette Coefficients for Robust Time Series Clustering

ICLR 2025 Conference Submission602 Authors

14 Sept 2024 (modified: 13 Oct 2024)ICLR 2025 Conference SubmissionEveryoneRevisionsBibTeXCC BY 4.0
Keywords: Data Mining, Time Series, Clustering
TL;DR: This paper introduces the first unsupervised time series clustering method that automatically determines the optimal number of clusters by applying the Silhouette Coefficient to SAX-based vector representations.
Abstract: Clustering algorithms are fundamental to data mining, serving dual roles as exploratory tools and preprocessing steps for advanced analytics. A persistent challenge in this domain is determining the optimal number of clusters, particularly for time series data where prevalent algorithms like k-means and k-shape require a priori knowledge of cluster quantity. This paper presents the first approach to time series clustering that does not require prior specification of cluster numbers. We introduce a novel extension of the Symbolic Pattern Forest (SPF) algorithm that automatically optimizes the number of clusters for time series datasets. Our method integrates SPF for cluster generation with the Silhouette Coefficient, computed on a two-stage vector representation: first transforming time series into Symbolic Aggregate approXimation (SAX) representations, then deriving both bag-of-words and TF-IDF vectors. Rigorous evaluation on diverse datasets from the UCR archive demonstrates that our approach significantly outperforms traditional baseline methods. This work contributes to the field of time series analysis by providing a truly unsupervised, data-driven approach to clustering, with potential impacts across various temporal data mining applications where the underlying number of clusters is unknown or variable.
Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 602
Loading