M-SpectraCast: Multivariate Cluster-Enhanced Time-Series Forecasting for Wireless Networks
IEEE Open Journal of the Communications Society · 2026 · 7 · 13525-13547
DOI: 10.1109/ojcoms.2026.3743620 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Open Journal of the Communications Society, 2026. Verfasst von Sadeq Bani Melhem und 3 weiteren Autor:innen. Die Arbeit nennt 29 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Natural Sciences and Engineering Research Council (NSERC) of Canada, European Union through Italian National Recovery and Resilience Plan (NRRP) of NextGenerationEU.
Abstract (Originalsprache)
Modern wireless networks require multivariate time-series forecasting for key tasks, such as traffic load prediction, spectrum analytics, signal strength assessments, and electromagnetic field exposure prediction. However, existing deep learning (DL) models for multivariate forecasting either overlook cross-channel dependencies through channel-independent (CI) designs or struggle to model heterogeneous interactions within fully channel-dependent (CD) frameworks. To address these challenges, we propose M-SpectraCast, a clustering-enhanced DL-enabled multivariate forecasting framework tailored for correlated multidimensional wireless time-series data. By grouping channels with similar statistical features, M-SpectraCast enables selective information sharing at the cluster level, effectively balancing CI and CD modeling while limiting dependency-modeling complexity and supporting robust forecasting across heterogeneous measurement settings. We conduct extensive experiments on large-scale multi-frequency wireless measurement datasets collected over multiple years in Italy, which we have released to support reproducibility and future research. In addition, we perform a comprehensive statistical and spectral characterization of the datasets through cross-frequency dependency analysis, temporal correlation analysis, stationarity assessment, and lag-based dynamics, uncovering structured inter-channel relationships across frequency bands. Experimental results across multiple RF-EMF datasets demonstrate that M-SpectraCast effectively exploits structured inter-channel dependencies, with the forecasting gains varying across datasets according to their underlying dependency and temporal structures. M-SpectraCast achieves up to 7.54% forecasting error reduction compared with the corresponding forecasting baselines, while maintaining competitive performance under cross-dataset evaluation. The results further show that competitive forecasting performance can be achieved with a small number of latent clusters.
Förderung
- Natural Sciences and Engineering Research Council (NSERC) of Canada
- European Union through Italian National Recovery and Resilience Plan (NRRP) of NextGenerationEU
Zitieren
APA
Melhem, S. B., Tabassum, H., Chiaraviglio, L., & Nguyen, U. T. (2026). M-SpectraCast: Multivariate Cluster-Enhanced Time-Series Forecasting for Wireless Networks. IEEE Open Journal of the Communications Society, 7, 13525-13547. https://doi.org/10.1109/ojcoms.2026.3743620
Vancouver
Melhem SB, Tabassum H, Chiaraviglio L, Nguyen UT. M-SpectraCast: Multivariate Cluster-Enhanced Time-Series Forecasting for Wireless Networks. IEEE Open Journal of the Communications Society. 2026;7:13525-13547. doi:10.1109/ojcoms.2026.3743620
BibTeX
@article{Melhem2026,
author = {Melhem, Sadeq Bani and Tabassum, Hina and Chiaraviglio, Luca and Nguyen, Uyen Trang},
title = {M-SpectraCast: Multivariate Cluster-Enhanced Time-Series Forecasting for Wireless Networks},
journal = {IEEE Open Journal of the Communications Society},
year = {2026},
volume = {7},
pages = {13525--13547},
doi = {10.1109/ojcoms.2026.3743620}
}
EUVIMED