EUVIMEDEUVIMED entdecken
Systematische Übersicht

Unidirectional and bidirectional long short-term memory for multiple seasonal time series data: A systematic review

Nur Haizum Abd Rahman, Roslinazairimah Zakaria, Nor Azuana Ramli

Multidisciplinary Reviews · 22.10.2026 · 10 (4) · 2027247

DOI: 10.31893/multirev.2027247 ↗

Kurz gefasst

Systematische Übersicht in der Zeitschrift Multidisciplinary Reviews, online erschienen am 22.10.2026. Verfasst von Nur Haizum Abd Rahman und 2 weiteren Autor:innen. Die Arbeit nennt 63 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This systematic review examines developments from January 2023 to November 2025 in unidirectional and bidirectional long short-term memory (LSTM and BiLSTM) models for forecasting multiple seasonal time series, a task complicated by overlapping cycles, nonlinear behaviors, and spatiotemporal variability across real-world datasets. While deep learning has become a dominant approach for handling complex seasonal structures, the rapid expansion of architecture variations, hybrid designs, and domain-specific implementations presents fragmentation and uncertainty regarding the most effective modeling strategies. To address this gap, the preferred reporting items for systematic reviews and meta-analyses (PRISMA) protocol was applied using two major databases, Web of Science and Scopus. From an initial pool of records, 45 published primary studies met all the inclusion criteria, offering a focused and contemporary evidence base. The selected studies were synthesized into three thematic clusters: (1) Renewable and power systems forecasting, where studies showed that seasonal–spatiotemporal learning with BiLSTM, decomposition techniques, and attention mechanisms improved the prediction of wind, solar, load, and hybrid energy systems; (2) Environmental quality and hydroclimate forecasting, where hybrid, feature selection, and spatial weighting enhanced the modeling of air pollution, streamflow, rainfall, evapotranspiration, and carbon-related emissions; and (3) Mobility, demand, and market forecasting, where multistage decomposition and hybrid recurrent networks outperformed traditional baselines in traffic flow, tourism demand, commodity prices, and agricultural yield prediction. Numerical evidence across themes demonstrated frequent reductions in forecasting error and improved long-range stability when BiLSTM or hybrid architectures were used. Hence, the review concludes that integrating decomposition, attention, feature engineering, and bidirectional recurrent structures offers a strong and adaptable foundation for multiseasonal forecasting while also highlighting remaining challenges in interpretability, generalizability, and computational efficiency.

Zitieren

APA

Rahman, N. H. A., Zakaria, R., & Ramli, N. A. (2026). Unidirectional and bidirectional long short-term memory for multiple seasonal time series data: A systematic review. Multidisciplinary Reviews, 10(4), 2027247. https://doi.org/10.31893/multirev.2027247

Vancouver

Rahman NHA, Zakaria R, Ramli NA. Unidirectional and bidirectional long short-term memory for multiple seasonal time series data: A systematic review. Multidisciplinary Reviews. 2026;10(4):2027247. doi:10.31893/multirev.2027247

BibTeX
@article{Rahman2026,
  author = {Rahman, Nur Haizum Abd and Zakaria, Roslinazairimah and Ramli, Nor Azuana},
  title = {Unidirectional and bidirectional long short-term memory for multiple seasonal time series data: A systematic review},
  journal = {Multidisciplinary Reviews},
  year = {2026},
  volume = {10},
  number = {4},
  pages = {2027247},
  doi = {10.31893/multirev.2027247}
}
Kostenlos starten →