Shapeleter

Shapelet Enhancer

---

1) Paper

Title: Shapeleter: Shapelet Enhancer via Multiview Positional Embedding for Time Series Classification
Venue: Knowledge-Based Systems (Under Review)

---

2) What Shapeleter does

Shapeleter is an interpretable shapelet-based TSC framework that targets both accuracy and interpretability.
It (i) selects a compact set of discriminative shapelets using a hypergraph-aware selection strategy to capture higher-order relations among candidates, (ii) represents selected shapelets with multiview position-aware features using two complementary notions of position to encode global context and ordering-related evolution, and (iii) performs prediction via a lightweight multiview ensemble built on simple linear classifiers.

Interpretability: The selected shapelets are human-readable subsequences and can be inspected per class as discriminative knowledge units.

---

Citation

If you use this repository (code, results, figures, or ideas), please cite our manuscript.

BibTeX

@unpublished{shapeleter_under_review,
  title   = {Shapeleter: Shapelet Enhancer via Multiview Positional Embedding for Time Series Classification},
  note    = {Under review at Knowledge-Based Systems},
  year    = {2026},
  url     = {https://github.com/CCHe64/Shapeleter}
}