EUVIMEDEUVIMED entdecken
Crossref

A Graph-Based Concept Drift Detection Method for Semi-Supervised Data Stream Classification

Shilun Ma, Hongliang Liu, Xinbo Jiang, Changyuan Chen

IEEE Signal Processing Letters · 2026 · 33 · 3054-3057

DOI: 10.1109/lsp.2026.3712202 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Signal Processing Letters, 2026. Verfasst von Shilun Ma und 3 weiteren Autor:innen. Die Arbeit nennt 15 Literaturangaben. Gefördert von: National Key R&D Program of China, National Natural Science Foundation of China, Taishan Scholars Program.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Semi-supervised data stream classification is challenging because only a small fraction of arriving samples are labeled, while the underlying data distribution may change over time. Existing concept drift detection methods mainly rely on classification errors or distributional discrepancies, which may be unreliable under limited-label conditions and often ignore local structural relationships among samples. To address this issue, this paper proposes SSCGD, a graph-structure-based concept drift detection method for semi-supervised data streams. SSCGD first constructs an optimized K-associated graph for each incoming data block by relaxing the subgraph merging criterion, thereby reducing fragmented subgraphs and better preserving local sample consistency. A historical classifier is then used to predict sample labels, and concept drift is detected according to the changes in edge weights and subgraph purity caused by inconsistent predictions. Experiments on synthetic and real-world data streams show that embedding SSCGD into existing semi-supervised classifiers improves average classification accuracy and enhances stability under low-label conditions.

Förderung

  • National Key R&D Program of China
  • National Natural Science Foundation of China
  • Taishan Scholars Program
  • Natural Science Foundation of Shandong Province

Zitieren

APA

Ma, S., Liu, H., Jiang, X., & Chen, C. (2026). A Graph-Based Concept Drift Detection Method for Semi-Supervised Data Stream Classification. IEEE Signal Processing Letters, 33, 3054-3057. https://doi.org/10.1109/lsp.2026.3712202

Vancouver

Ma S, Liu H, Jiang X, Chen C. A Graph-Based Concept Drift Detection Method for Semi-Supervised Data Stream Classification. IEEE Signal Processing Letters. 2026;33:3054-3057. doi:10.1109/lsp.2026.3712202

BibTeX
@article{Ma2026,
  author = {Ma, Shilun and Liu, Hongliang and Jiang, Xinbo and Chen, Changyuan},
  title = {A Graph-Based Concept Drift Detection Method for Semi-Supervised Data Stream Classification},
  journal = {IEEE Signal Processing Letters},
  year = {2026},
  volume = {33},
  pages = {3054--3057},
  doi = {10.1109/lsp.2026.3712202}
}
Kostenlos starten →