INSHAPE: Interpretable Time-Series Classification via Instance-Level Shapelets

A new research paper introduces INSHAPE, a framework for interpretable time-series classification (TSC) that discovers variable-length, discriminative temporal patterns specific to each individual time series. The paper, available on arXiv (2605.20088), addresses limitations of existing shapelet methods, which typically focus on population-level patterns optimized across an entire dataset. These population-level patterns often misalign with instance-specific features, leading to suboptimal performance and potentially misleading interpretations. Additionally, most methods treat shapelets as independent entities, overlooking temporal dependencies and interactions among multiple patterns. INSHAPE identifies non-overlapping segments as shapelets, tailored to each time series, enhancing interpretability and performance. The framework is designed to make model decision-making processes more transparent. The paper is a preprint, marked as 'replace-cross', indicating a revision. The authors propose INSHAPE as a solution to the inherent complexity of TSC, aiming to provide more accurate and interpretable models. The research is relevant to the fields of machine learning and data mining, with potential applications in various domains where time-series data are prevalent, such as finance, healthcare, and sensor analytics. The paper does not specify a particular institution or author names in the provided content, but it is hosted on arXiv, a repository for scientific preprints.