EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

SHAP-LM: Optimized Music Genre Classification via SHAP-Guided Feature Selection and Lightweight Modeling

Yinghui Wang, Jisen Jiang, Mengqi Zhang, Huasen Zhang

IEEE Access · 2026 · 14 · 27733-27742

DOI: 10.1109/access.2026.3664687 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Yinghui Wang und 3 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 44 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Music genre classification (MGC) is a fundamental task in music information retrieval, yet existing approaches often suffer from high computational cost, limited generalization, and lack of interpretability. To address these challenges, we propose SHAP-LM, a lightweight and interpretable framework that integrates SHapley Additive exPlanations (SHAP)-guided feature selection with a hybrid architecture combining LightGBM and a parameter-efficient deep neural network. By retaining only the most informative acoustic attributes, the model reduces feature dimensionality by 75% while preserving discriminative power. On the GTZAN dataset, SHAP-LM achieves 97.1% classification accuracy with an inference latency of 42 ms, outperforming strong baselines in both accuracy and efficiency. Furthermore, the mapping of SHAP values back to raw waveforms enhances interpretability, offering insights into genre-specific acoustic cues. Experimental results demonstrate that our method provides a scalable, transparent, and resource-efficient solution for real-world MGC applications.

Zitieren

APA

Wang, Y., Jiang, J., Zhang, M., & Zhang, H. (2026). SHAP-LM: Optimized Music Genre Classification via SHAP-Guided Feature Selection and Lightweight Modeling. IEEE Access, 14, 27733-27742. https://doi.org/10.1109/access.2026.3664687

Vancouver

Wang Y, Jiang J, Zhang M, Zhang H. SHAP-LM: Optimized Music Genre Classification via SHAP-Guided Feature Selection and Lightweight Modeling. IEEE Access. 2026;14:27733-27742. doi:10.1109/access.2026.3664687

BibTeX
@article{Wang2026,
  author = {Wang, Yinghui and Jiang, Jisen and Zhang, Mengqi and Zhang, Huasen},
  title = {SHAP-LM: Optimized Music Genre Classification via SHAP-Guided Feature Selection and Lightweight Modeling},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {27733--27742},
  doi = {10.1109/access.2026.3664687}
}
Kostenlos starten →