Islam et al. (2025) From Traditional Machine Learning to Fine-Tuning Large Language Models: A Review for Sensors-Based Soil Moisture Forecasting

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Sensors
• Year: 2025
• Date: 2025-11-12
• Authors: Md Babul Islam, Antonio Guerrieri, Raffaele Gravina, Declan Delaney, Giancarlo Fortino
• DOI: 10.3390/s25226903

Research Groups

This paper is a systematic literature review and does not involve primary experimental or modeling research groups in the traditional sense.

Short Summary

This paper proposes a novel taxonomy for soil moisture (SM) forecasting and provides a comprehensive review of 68 peer-reviewed studies published between 2017 and 2025, covering traditional machine learning, deep learning, and hybrid models, while identifying future research directions.

Objective

• To propose a novel taxonomy for Soil Moisture (SM) forecasting and provide a comprehensive review of existing approaches, including recent advancements like Federated Learning (FL), Transfer Learning (TL), and Large Language Models (LLMs), to address gaps in current literature.

Study Configuration

• Spatial Scale: Not applicable for the review itself, as it synthesizes literature across various spatial scales addressed by the reviewed studies.
• Temporal Scale: The review covers studies published between 2017 and 2025. The reviewed studies themselves address various temporal scales for soil moisture forecasting.

Methodology and Data

• Models used: The review itself uses the PRISMA methodology. The reviewed papers utilize traditional machine learning, deep learning, and hybrid models, including recent advancements like Federated Learning (FL), Transfer Learning (TL), and Large Language Models (LLMs).
• Data sources: The review's data source is 68 peer-reviewed scientific papers published between 2017 and 2025. The reviewed papers primarily use data from soil sensors, local weather data, and other real-time sensing systems.

Main Results

• A novel taxonomy for Soil Moisture (SM) forecasting is proposed.
• 68 peer-reviewed studies published between 2017 and 2025 were systematically analyzed based on sensor types, input features, AI techniques, data durations, and evaluation metrics.
• Promising research directions are identified, including the application of TinyML for edge deployment, explainable AI for improved transparency, and privacy-aware model training.

Contributions

• Proposes a novel, structured taxonomy for SM forecasting, addressing a gap in existing review articles.
• Provides a comprehensive review incorporating recent advancements (e.g., Federated Learning, Transfer Learning, Large Language Models) often overlooked in previous literature.
• Identifies future research directions to guide the development of accurate, scalable, and trustworthy SM forecasting systems for Smart Agriculture.
• Utilizes the PRISMA methodology for a systematic and rigorous review process.

Funding

• Not specified in the provided text.

Citation

@article{Islam2025From,
  author = {Islam, Md Babul and Guerrieri, Antonio and Gravina, Raffaele and Delaney, Declan and Fortino, Giancarlo},
  title = {From Traditional Machine Learning to Fine-Tuning Large Language Models: A Review for Sensors-Based Soil Moisture Forecasting},
  journal = {Sensors},
  year = {2025},
  doi = {10.3390/s25226903},
  url = {https://doi.org/10.3390/s25226903}
}